# Louis-François Bouchard, aka What's AI Complete Public Content Index

> Making AI accessible through practical AI engineering writing, videos, courses, and consulting.

This is a complete Markdown index of canonical public URLs, designed for assistants and agents. It intentionally avoids duplicating every article body so canonical HTML pages remain the primary source for indexing, sharing, and attribution.

## Discovery Files
- [AI index](https://www.louisbouchard.ai/ai-index.json): Structured JSON with topics, tags, takeaways, FAQs, key questions, source links, media, and reviewed dates.
- [Image sitemap](https://www.louisbouchard.ai/image-sitemap.xml): Local images mapped to public URLs.
- [Video sitemap](https://www.louisbouchard.ai/video-sitemap.xml): YouTube embeds mapped to canonical pages.
- [RSS](https://www.louisbouchard.ai/rss.xml): Recent articles.
- [JSON Feed](https://www.louisbouchard.ai/feed.json): Recent articles in JSON Feed format.

## Pages
- [About me](https://www.louisbouchard.ai/about/): Learn who Louis-François Bouchard is, how What's AI started, and how his work connects AI engineering, education, videos, and Towards AI. Updated Jul 09, 2026.
- [We Can Help your Team Adopt AI: Corporate Training, Consulting, and Talent Solutions](https://www.louisbouchard.ai/b2b/): AI training, consulting, and talent support for teams that want to adopt AI with practical systems, not vague demos or hype. Updated May 22, 2024.
- [Contact](https://www.louisbouchard.ai/contact/): Contact Louis-François Bouchard for AI training, consulting, sponsorships, podcast requests, collaborations, or practical AI engineering questions. Updated Jul 10, 2026.
- [FAQ](https://www.louisbouchard.ai/faq/): Answers about Louis-François Bouchard, What's AI, Towards AI, courses, sponsorships, B2B training, speaking, and the community. Updated Jul 09, 2026.
- [From zero to hero with LLMs](https://www.louisbouchard.ai/from-zero-to-hero-with-llms/): Learn large language models in 2026 with a practical path through prerequisites, transformers, prompting, RAG, fine-tuning, projects, and free courses. Updated Jan 23, 2026. Audio: [original video audio](https://www.louisbouchard.ai/audio/from-zero-to-hero-with-llms-58XmJxb1x_o-v1.mp3), sourced from [Get Started with LLMs. Work with Large Language Models in 2024](https://www.youtube.com/watch?v=58XmJxb1x_o).
- [Learn AI Engineering](https://www.louisbouchard.ai/learn-ai-engineering/): Follow a practical 2026 AI engineering roadmap from Python and LLM basics through RAG, agents, evals, deployment, security, projects, and job preparation. Updated Aug 03, 2026. Audio: [original video audio](https://www.louisbouchard.ai/audio/learn-ai-engineering-8spm9E2RsXI-v1.mp3), sourced from [How I'd Learn AI Engineering in 2026](https://www.youtube.com/watch?v=8spm9E2RsXI).
- [Learn AI Together](https://www.louisbouchard.ai/learn-ai-together/): Join the Learn AI Together community to learn AI with other builders, ask questions, share projects, and stay close to practical AI work. Updated Feb 13, 2026.
- [Start AI in 2026 — Become an expert from nothing, for free!](https://www.louisbouchard.ai/learnai/): Learn AI and machine learning from scratch in 2026 with a free roadmap for coding, math, courses, projects, communities, portfolios, and your first job. Updated Jan 23, 2026. Audio: [original video audio](https://www.louisbouchard.ai/audio/learnai-RirEw-uaS_8-v1.mp3), sourced from [Learn AI for free in 2024](https://www.youtube.com/watch?v=RirEw-uaS_8).
- [Louis-François Bouchard Weekly Newsletter](https://www.louisbouchard.ai/newsletter/): Get practical AI engineering notes, research explanations, useful tools, and new What's AI videos in Louis-François Bouchard's newsletter on Substack. Updated Nov 16, 2023.
- [The What's AI Podcast by Louis-François Bouchard](https://www.louisbouchard.ai/podcast/): The What's AI Podcast features conversations with builders, researchers, founders, and educators working on practical AI and machine learning. Updated Dec 31, 2023.
- [Privacy](https://www.louisbouchard.ai/privacy/): How louisbouchard.ai handles analytics, A/B experiments, contact messages, comments, AI helper conversations, browser settings, and third-party embeds. Updated Aug 05, 2026.
- [My Sponsors](https://www.louisbouchard.ai/sponsors/): Sponsor What's AI to reach builders, developers, students, and teams who care about practical AI education and AI engineering. Updated Nov 18, 2023.
- [My videos](https://www.louisbouchard.ai/videos/): Watch What's AI on YouTube for practical explanations of AI engineering, LLMs, agents, computer vision, research papers, and tools for builders and learners. Updated May 22, 2024.

## Topic Hubs
- [AI Agents](https://www.louisbouchard.ai/topics/ai-agents/): Practical writing on AI agents, workflows, memory, tool use, reliability, and what actually ships. 31 articles.
- [Large Language Models](https://www.louisbouchard.ai/topics/llms/): Clear explanations of LLMs, reasoning models, prompting, context, distillation, fine-tuning, and model behavior. 103 articles.
- [RAG and Retrieval](https://www.louisbouchard.ai/topics/rag/): Guides and explainers on retrieval augmented generation, vector databases, embeddings, indexing, and evaluation. 26 articles.
- [AI Engineering](https://www.louisbouchard.ai/topics/ai-engineering/): Roadmaps, architecture decisions, evals, deployment lessons, and practical AI engineering advice. 78 articles.
- [Computer Vision](https://www.louisbouchard.ai/topics/computer-vision/): Computer vision explainers on image generation, segmentation, 3D reconstruction, video, and visual AI research. 94 articles.
- [Generative AI](https://www.louisbouchard.ai/topics/generative-ai/): How generative AI models create images, video, audio, code, and text, without the hype fog. 74 articles.
- [Learning AI](https://www.louisbouchard.ai/topics/learning-ai/): Free guides, beginner roadmaps, course recommendations, and practical paths into AI engineering. 54 articles.
- [AI News and Analysis](https://www.louisbouchard.ai/topics/ai-news-analysis/): Opinionated but fair analysis of AI releases, model news, research updates, and industry shifts. 73 articles.

## Posts (265)
- [DeepSeek V4.1 Flash: Smaller KV Cache, Cheaper Writing Drafts](https://www.louisbouchard.ai/deepseek-v41-flash-kv-cache-writing-benchmark/) - Published Sep 13, 2026. Updated Sep 13, 2026. Tags: AI Engineering, DeepSeek, KV Cache, LLM Evaluation. How DeepSeek V4.1 Flash reduces KV cache memory, what changed over V4 Flash, and what our writing benchmark measured for quality, cost, and effort. Questions: What is new in DeepSeek V4.1 Flash? Does DeepSeek V4.1 Flash fit in 1 GB of memory? Should I use high or max reasoning effort? Takeaways: V4.1’s global KV cache is about one quarter of V4 Flash’s; this is not total model memory. Default effort averaged 43 seconds and $0.007744 per writing draft in our September 10 snapshot. Audio: [original video audio](https://www.louisbouchard.ai/audio/deepseek-v41-flash-kv-cache-writing-benchmark--28P_Cvvg_w-v1.mp3), sourced from [How DeepSeek V4.1 Flash Actually Works: A Deep Dive](https://www.youtube.com/watch?v=-28P_Cvvg_w).
- [My AI Setup: Claude Code, Codex, Obsidian and a Mac Mini](https://www.louisbouchard.ai/my-ai-setup/) - Published Sep 09, 2026. Updated Sep 09, 2026. Tags: AI Engineering, Claude Code, Codex, Obsidian. Inside my personal AI setup: shared notes and skills, Claude Code and Codex, remote access from my phone, and scheduled work on an always-on Mac mini. Questions: Do I need Obsidian to use this approach? Does my laptop have to stay open? Can Claude Code and Codex share all their memory? Takeaways: Keep reusable knowledge in files your different agents can reach. A separate awake host lets remote sessions and scheduled work continue when your laptop closes. Audio: [original video audio](https://www.louisbouchard.ai/audio/my-ai-setup-KeyqBiQYMok-v1.mp3), sourced from [Why I Stopped Letting Codex Lead My Long Tasks](https://www.youtube.com/watch?v=KeyqBiQYMok).
- [How We Actually Use AI to Ship Code](https://www.louisbouchard.ai/ai-coding-workflows-that-ship/) - Published Aug 21, 2026. Updated Aug 21, 2026. Tags: AI Engineering, Coding Agents, Software Engineering. The three AI coding workflows, context rules, debugging habits, tests, and review guardrails the Towards AI engineering team uses to ship reliable code. Questions: What is the core rule for coding with AI agents? What context should I give a coding agent? When should I use a strong model instead of a fast model? Takeaways: Human judgment should decide the architecture, constraints, and acceptance criteria. Coding agents are most useful when they handle the typing and execution. Push lean project rules into CLAUDE.md or AGENTS.md, then pull only the files, logs, and task context needed for the current change. Audio: [original video audio](https://www.louisbouchard.ai/audio/ai-coding-workflows-that-ship-9SCkcVKPEy0-v1.mp3), sourced from [How I Actually Use Coding Agents](https://www.youtube.com/watch?v=9SCkcVKPEy0).
- [Context Engineering in 2026: Why We Stopped Compacting Our Agent's Context](https://www.louisbouchard.ai/context-engineering-2026/) - Published Aug 18, 2026. Updated Aug 18, 2026. Tags: AI Engineering, AI Agents, Large Language Models. Our AI tutor tests compare compaction and prompt caching: when keeping the full context costs less than summarizing it. Questions: What is context engineering? Should I compact or summarize my agent's chat history? Why does prompt caching make summarization expensive? Takeaways: Under modern prompt caching, keeping the full history beat every summarization strategy we tested on cost, latency, and memory recall at once. Summarizing rewrites the cached prefix, so you pay full price to recompute everything you just tried to save. The cheap tier of compaction is the part that pays. Capping every tool output at a stable size cut our cost per turn by 38% with no measurable loss in memory, because it shrinks the context without rewriting the prefix the cache depends on.
- [The Towards AI Mentorship for AI Engineers](https://www.louisbouchard.ai/towards-ai-mentorship-for-ai-engineers/) - Published Aug 04, 2026. Updated Aug 04, 2026. Tags: AI Engineering, Learning AI, Careers in AI. Meet the Towards AI Mentorship: senior engineering help for your AI projects, architecture, career, and path into production. Questions: What is the Towards AI Mentorship? Who is the Towards AI Mentorship for? Is the mentorship one-to-one coaching? Takeaways: The Towards AI Mentorship gives individual AI engineers direct access to a team of senior AI engineers for architecture, implementation, and career questions. Members receive two live sessions each week, asynchronous support, production resources, career feedback, community access, and monthly industry events. Audio: [original video audio](https://www.louisbouchard.ai/audio/towards-ai-mentorship-for-ai-engineers-hWObg9tj-ng-v1.mp3), sourced from [The Towards AI Mentorship for AI Engineers](https://www.youtube.com/watch?v=hWObg9tj-ng).
- [How I'd Learn AI Engineering in 2026](https://www.louisbouchard.ai/how-to-learn-ai-engineering-2026/) - Published Aug 03, 2026. Updated Aug 03, 2026. Tags: AI Engineering, Large Language Models, Learning AI. Learn AI engineering in 2026 through LLM foundations, RAG, agents, evals, and projects that build practical system-design judgment. Questions: What should I learn first to become an AI engineer in 2026? Are Codex, Claude Code, or Cursor enough to become an AI engineer? When should I use RAG, fine-tuning, a workflow, or an agent? Takeaways: Coding agents accelerate implementation, but AI engineers create value by choosing architectures, evaluating behavior, and owning failures. Learn the foundations quickly, then build small systems with retrieval, tools, structured outputs, and visible failure modes. Audio: [original video audio](https://www.louisbouchard.ai/audio/how-to-learn-ai-engineering-2026-8spm9E2RsXI-v1.mp3), sourced from [How I'd Learn AI Engineering in 2026](https://www.youtube.com/watch?v=8spm9E2RsXI).
- [How I Automated 90% of My Content Workflow With AI Agents](https://www.louisbouchard.ai/ai-agent-content-pipeline/) - Published Jul 28, 2026. Updated Aug 03, 2026. Tags: AI Agents, AI Engineering, Workflow Automation. A look inside the AI agent pipeline I use to research, write, illustrate, package, translate, and publish content while keeping the critical 10% human. Questions: What does the AI agent content pipeline automate? Why is research an agent but writing a workflow? How does the pipeline reduce hallucinations and silent failures? Takeaways: Automate research, drafting, media production, packaging, and translation, but keep topic selection, direction, taste, and final validation human. Use an exploratory agent for open-ended research and a deterministic, checkpointed workflow for writing where repeatability matters more than autonomy. Audio: [original video audio](https://www.louisbouchard.ai/audio/ai-agent-content-pipeline-psIh5S1KEMA-v1.mp3), sourced from [I Automated 90% of My YouTube Workflow](https://www.youtube.com/watch?v=psIh5S1KEMA).
- [Graph Engineering vs Loop Engineering: What Actually Changed](https://www.louisbouchard.ai/graph-engineering-explained/) - Published Jul 22, 2026. Updated Jul 22, 2026. Tags: AI Agents, AI Engineering, Large Language Models. Graph engineering explained: why the viral loops-versus-graphs debate misses the point, what actually changed in agent orchestration, and what should stay code. Questions: What is graph engineering? Is loop engineering dead? What is the difference between a loop and a graph of agents? Takeaways: Loop engineering is not dead. A graph is what you get when one loop is no longer enough, and graphs contain loops. The structure is old: workflow engines and DAG schedulers have drawn these graphs for a decade. What changed is that the nodes now interpret their tasks. Audio: [original video audio](https://www.louisbouchard.ai/audio/graph-engineering-explained-MDHcmWmvqvo-v1.mp3), sourced from [Graph Engineering, Without the Hype](https://www.youtube.com/watch?v=MDHcmWmvqvo).
- [How to Build a Memory Your AI Agents Can Actually Reuse](https://www.louisbouchard.ai/memory-for-ai-agents/) - Published Jun 26, 2026. Updated Jun 26, 2026. Tags: AI Agents, Large Language Models. Build reusable memory for AI agents with plain files, source references, an index, and a wiki that future sessions can inspect, query, and update. Questions: How do you build reusable memory for AI agents? Why not paste everything into the model context? What should agent memory avoid? Takeaways: The bottleneck is not giving the model more information once. It is making useful information available again later. Agent memory works better as external files and references than as one giant prompt pasted into every session.
- [What Did Microsoft's MAI Model Actually Train On?](https://www.louisbouchard.ai/mai-thinking/) - Published Jun 23, 2026. Updated Jun 23, 2026. Tags: AI News and Analysis. Microsoft MAI-Thinking-1 avoids third-party model outputs in pretraining. Its trillion-parameter MoE design and report show what it actually trained on. Questions: What Did Microsoft's MAI Model Actually Train On? What did Microsoft's MAI model train on? Why does training data matter here? Takeaways: The interesting MAI claim is not only benchmark performance. It is the choice to remove AI-generated data from training. If a model copies another model's reasoning, you inherit behavior that may be harder to audit, steer, or explain. Audio: [original video audio](https://www.louisbouchard.ai/audio/mai-thinking-Sl5O7KVVF6M-v1.mp3), sourced from [Microsoft Says MAI-Thinking-1 Was Not Distilled From Another LLM](https://www.youtube.com/watch?v=Sl5O7KVVF6M).
- [90% Cheaper GPT APIs](https://www.louisbouchard.ai/api-proxy/) - Published Jun 15, 2026. Updated Jun 15, 2026. Tags: Large Language Models. Cheap GPT API proxies can swap models, log keys and code, or inject malicious output. A 90% token discount can put your entire repository at risk. Questions: Why are cheap GPT API proxies risky? Can a proxy change the model you paid for? What should builders do before using one? Takeaways: A cheap API proxy is not free efficiency. Someone still pays for the tokens, the infrastructure, or the risk. The proxy can quietly swap the model, log secrets, inject code, or sit between your coding agent and your repo. Audio: [original video audio](https://www.louisbouchard.ai/audio/api-proxy-sMvIMDWVLmQ-v1.mp3), sourced from [The hidden costs of cheap AI APIs](https://www.youtube.com/watch?v=sMvIMDWVLmQ).
- [Loop Engineering Explained](https://www.louisbouchard.ai/loop-engineering/) - Published Jun 10, 2026. Updated Jun 10, 2026. Tags: AI Agents, Large Language Models, AI Engineering. Loop engineering for coding agents replaces babysitting with repeatable workflows for context, tools, checks, budgets, recovery, and clear stop conditions. Questions: What is loop engineering? How is this different from prompting? What can go wrong with loops? Takeaways: The point is not to prompt the agent harder. It is to design the loop the agent runs inside. A good loop defines context, tools, success criteria, failure handling, budget, and when to stop. Audio: [original video audio](https://www.louisbouchard.ai/audio/loop-engineering-NjXIIH9vcv0-v1.mp3), sourced from [Loop Engineering Explained](https://www.youtube.com/watch?v=NjXIIH9vcv0).
- [How Image Generation Actually Works](https://www.louisbouchard.ai/image-generation/) - Published Jun 07, 2026. Updated Jun 07, 2026. Tags: Large Language Models, Computer Vision, Generative AI. How AI image generation works: diffusion sculpts noise, autoregressive models predict visual tokens, and text prompts steer both toward the final pixels. Questions: How does AI image generation work? Why do prompts change the image? What should users be careful about? Takeaways: Image models do not draw from words directly. They learn a visual space where prompts guide what kind of image fits. The prompt controls constraints, but the model still fills in details from patterns learned during training. Audio: [original video audio](https://www.louisbouchard.ai/audio/image-generation-DCOXcnOntmg-v1.mp3), sourced from [How AI Image Generation Actually Works (There Are Only 2 Ways)](https://www.youtube.com/watch?v=DCOXcnOntmg).
- [Coding Changed Forever](https://www.louisbouchard.ai/vibe-coding/) - Published May 25, 2026. Updated May 25, 2026. Tags: AI Agents, Large Language Models, AI News and Analysis. Vibe coding with AI agents can speed up software work, but only with good context, tests, reviews, and task boundaries. See what changed and where it fails. Questions: What is vibe coding? How do you use coding agents safely? What is the danger of vibe coding? Takeaways: AI can write a lot of code, but that does not remove the need for engineering judgment. Good agentic coding depends on context files, tests, skills, code review, and clear task boundaries. Audio: [original video audio](https://www.louisbouchard.ai/audio/vibe-coding-ShFn3MG0h8s-v1.mp3), sourced from [Vibe Coding: The Complete Story (2015–2026)](https://www.youtube.com/watch?v=ShFn3MG0h8s).
- [How Google Went From AI Joke to OpenAI’s Biggest Problem](https://www.louisbouchard.ai/googleai/) - Published May 01, 2026. Updated May 01, 2026. Tags: Large Language Models, AI News and Analysis. Google AI's comeback explained: how Gemini, DeepMind, AI Studio, and product distribution turned a mocked chatbot launch into a serious OpenAI rival. Questions: Why did people underestimate Google in AI? What changed for Google AI? What should builders take from this? Takeaways: Google was easy to underestimate, but its research base, infrastructure, and product surface never disappeared. The interesting shift is not one model release. It is how fast Google turned scattered AI pieces into a stronger product story. Audio: [original video audio](https://www.louisbouchard.ai/audio/googleai-4hkY-58M3hw-v1.mp3), sourced from [Why Google May Win the Next Phase of AI](https://www.youtube.com/watch?v=4hkY-58M3hw).
- [Your AI Can Improve Itself — Or Fool You](https://www.louisbouchard.ai/self-improvement/) - Published Apr 24, 2026. Updated Apr 24, 2026. Tags: Computer Vision. Recursive self-improvement in AI uses generate, test, and keep-or-discard loops. Strong evals separate real progress from a system that learns to fool you. Questions: What does self-improving AI mean here? Why can self-improvement fool you? How should builders use self-improving loops? Takeaways: Self-improvement is often a set of small loops: propose a change, test it, keep what works, and repeat. The scary part is not only autonomy. It is a system improving against a bad reward or weak evaluation.
- [AI Distillation Explained: The Truth Behind the Biggest AI Controversy Right Now](https://www.louisbouchard.ai/ai-distillation/) - Published Apr 17, 2026. Updated Apr 17, 2026. Tags: Large Language Models. AI model distillation explained: how student models copy a teacher's behavior, why labs use it, and where data rights, API terms, and ownership collide. Questions: What is AI distillation? Why is AI distillation controversial? What should builders learn from this debate? Takeaways: Distillation is technically simple: use a stronger model's outputs to train a smaller or cheaper model. The controversy is not only about capability copying. It is about data rights, API terms, and how labs prove what they trained on. Audio: [original video audio](https://www.louisbouchard.ai/audio/ai-distillation-BkhL4Az100U-v1.mp3), sourced from [Who's Really Stealing From Whom?](https://www.youtube.com/watch?v=BkhL4Az100U).
- [How to Control Bias in AI Agents](https://www.louisbouchard.ai/control-bias-in-ai-agents/) - Published Apr 01, 2026. Updated Apr 01, 2026. Tags: AI Agents, AI News and Analysis. Control bias in AI agents with system-level audits, scoped tools, feedback loops, and human review. See how autonomy can compound a small skew over time. Questions: Why does bias get harder with AI agents? How can teams control agent bias? What is the mistake to avoid? Takeaways: More autonomy can amplify bias if the agent has tools, memory, and planning without the right checks around it. Bias is not always a defect. Some preferences are intended, but they need to be explicit, tested, and monitored. Audio: [original video audio](https://www.louisbouchard.ai/audio/control-bias-in-ai-agents---WihC6AJOw-v1.mp3), sourced from [Will AI Agents Make Bias Worse?](https://www.youtube.com/watch?v=--WihC6AJOw).
- [Harness Engineering: The Missing Layer Behind AI Agents](https://www.louisbouchard.ai/harness-engineering/) - Published Mar 25, 2026. Updated Mar 25, 2026. Tags: AI Agents, Large Language Models, AI Engineering. Harness engineering explained: the tools, permissions, traces, evals, guardrails, and approvals that make AI agents debuggable and safer to ship. Questions: What is harness engineering for AI agents? Why does an agent need a harness? What happens without a harness? Takeaways: An agent without a harness is hard to debug because you cannot reliably see, replay, or constrain what happened. The harness is where tools, permissions, evaluation, logging, and human approvals turn a demo into a system. Audio: [original video audio](https://www.louisbouchard.ai/audio/harness-engineering-zYerCzIexCg-v1.mp3), sourced from [What Harness Engineering Actually Means](https://www.youtube.com/watch?v=zYerCzIexCg).
- [Why RAG Is Not Training Your AI](https://www.louisbouchard.ai/why-rag-is-not-training-your-ai/) - Published Mar 16, 2026. Updated Mar 16, 2026. Tags: Large Language Models, RAG and Retrieval. RAG gives an AI model temporary context without changing its weights. See how retrieval differs from training and when fine-tuning is the better tool. Questions: Is RAG the same as training an AI model? What does RAG actually change? Why does this distinction matter? Takeaways: RAG adds information at query time. It does not rewrite the model's internal weights or permanently teach it your data. That distinction matters for privacy, expectations, debugging, and how teams talk about AI knowledge. Audio: [original video audio](https://www.louisbouchard.ai/audio/why-rag-is-not-training-your-ai-zD7ntSjTwVA-v1.mp3), sourced from [No, Pasting Data Into ChatGPT Does Not Train It](https://www.youtube.com/watch?v=zD7ntSjTwVA).
- [How to Get an AI Engineer Job in 2026](https://www.louisbouchard.ai/what-i-look-for-when-hiring-ai-engineers/) - Published Mar 09, 2026. Updated Aug 28, 2026. Tags: AI Engineering. Prepare for AI engineering interviews in 2026, build a portfolio that earns callbacks, and use coding agents responsibly. Questions: How should I prepare for an AI engineering interview in 2026? What should be in an AI engineer portfolio? Can I use coding agents in a take-home interview? Takeaways: Ship a small AI product end to end instead of stopping at a notebook or generic demo. Use a verified test set, a baseline, quantitative results, and error analysis to prove it works. Audio: [original video audio](https://www.louisbouchard.ai/audio/what-i-look-for-when-hiring-ai-engineers-Z1O0CsljJWo-v1.mp3), sourced from [AI Engineer Skills You Need to Get Hired in 2026](https://www.youtube.com/watch?v=Z1O0CsljJWo).
- [Stop Building Agent Demos](https://www.louisbouchard.ai/stop-building-agent-demos/) - Published Feb 26, 2026. Updated Feb 26, 2026. Tags: AI Agents, AI Engineering. Move from AI agent demos to systems you can evaluate, debug, deploy, and trust, with scoped tools, traces, evals, failure handling, and real users. Questions: Why stop building agent demos? What makes an agent production-ready? What is the agent demo trap? Takeaways: Agent demos are easy to make impressive, but production systems need reliability, traces, evals, and failure handling. The hard question after a local demo is: how do you know it worked, why it failed, and what changed after a fix? Audio: [original video audio](https://www.louisbouchard.ai/audio/stop-building-agent-demos-Y5_NMQgV0Z8-v1.mp3), sourced from [Our New Agent Engineering Course!](https://www.youtube.com/watch?v=Y5_NMQgV0Z8).
- [I Published 42 Shorts on AI Terms in 42 days](https://www.louisbouchard.ai/ai-in-42-terms/) - Published Feb 05, 2026. Updated Feb 10, 2026. Tags: Large Language Models. Understand 42 AI and LLM terms through a free YouTube course of short videos on tokens, context, model behavior, and why fluent answers can still fail. Questions: Why make 42 short videos about AI terms? What does this teach beyond prompting? What mistake does this project push against? Takeaways: Prompting was never the whole problem. You need to understand what kind of system you are talking to. AI tools can sound confident while failing simple reasoning, grounding, or consistency checks. Audio: [original video audio](https://www.louisbouchard.ai/audio/ai-in-42-terms-TnpU6SSBWao-v1.mp3), sourced from [42 AI Concepts You Actually Need to Understand LLMs](https://www.youtube.com/watch?v=TnpU6SSBWao).
- [The 12 Questions That Decide Your AI Architecture](https://www.louisbouchard.ai/12-questions-ai-architecture/) - Published Jan 29, 2026. Updated Jan 29, 2026. Tags: AI Agents, AI Engineering. A practical checklist for choosing AI architecture before implementation, so teams avoid agent rewrites caused by bad scoping. Questions: What do the 12 AI architecture questions help decide? When is a single AI agent still enough? What is the expensive mistake this avoids? Takeaways: Architecture problems usually start before code, when teams choose agents, workflows, or frameworks before defining the job. Scope comes first: what the system must decide, what tools it can use, and where humans still need to approve work. Audio: [original video audio](https://www.louisbouchard.ai/audio/12-questions-ai-architecture-CMMlLB01rcE-v1.mp3), sourced from [The 12 Questions That Decide Your AI Architecture](https://www.youtube.com/watch?v=CMMlLB01rcE).
- [Multi-agent is becoming the new overengineering](https://www.louisbouchard.ai/agents-and-workflows/) - Published Jan 28, 2026. Updated Jan 28, 2026. Tags: AI Agents, AI Engineering. A clear distinction between workflows, agents, and multi-agent systems, with the main rule: stay as simple as the problem allows. Questions: What is the difference between a workflow and an agent? When should you use a multi-agent system? What is the common overengineering mistake? Takeaways: Most things called agents are still workflows, and that is fine if the steps are known in advance. Autonomy is the key difference: in a workflow you control the steps, while an agent decides what to do next. Audio: [original video audio](https://www.louisbouchard.ai/audio/agents-and-workflows-_rO2fv6tSsQ-v1.mp3), sourced from [Stop Overengineering: Workflows vs AI Agents Explained](https://www.youtube.com/watch?v=_rO2fv6tSsQ).
- [How to Clean Up AI-Generated Drafts Without Sounding Like ChatGPT](https://www.louisbouchard.ai/ai-editing/) - Published Jan 15, 2026. Updated Jan 15, 2026. Tags: AI Agents, Large Language Models, Generative AI. A practical editing workflow for using LLMs without losing your voice, with checks for slop, repetition, and fake polish. Questions: How do you clean up AI-generated writing? What makes AI writing sound like ChatGPT? What should you avoid when editing with AI? Takeaways: AI writing usually fails in predictable ways: generic words, repeated structure, over-polished claims, and weak human judgment. A good editing loop freezes the draft, judges it against evidence, then rewrites the weak parts once with a clear plan. Audio: [original video audio](https://www.louisbouchard.ai/audio/ai-editing-bMlvTbg1hVk-v1.mp3), sourced from [5 Edits That Instantly Make AI Text Sound Human](https://www.youtube.com/watch?v=bMlvTbg1hVk).
- [Why AI Feels Intelligent (and Why That’s Misleading)](https://www.louisbouchard.ai/why-ai-feels-intelligent/) - Published Dec 23, 2025. Updated Dec 23, 2025. Tags: Large Language Models, Learning AI. Why LLMs feel intelligent, why the comparison to human learning gets messy, and what builders should remember about fluency. Questions: Why does AI feel intelligent? Why is the human comparison misleading? What should builders remember? Takeaways: LLMs and humans both learn patterns, but the signals, grounding, and relationship to the world are very different. A model can imitate language impressively without having human experience, goals, or understanding behind the words. Audio: [original video audio](https://www.louisbouchard.ai/audio/why-ai-feels-intelligent-Pm4P6ryfezI-v1.mp3), sourced from [Prediction Isn’t Understanding and That Difference Matters](https://www.youtube.com/watch?v=Pm4P6ryfezI).
- [Earn While Helping Others Learn AI](https://www.louisbouchard.ai/affiliate/) - Published Sep 02, 2025. Updated Sep 02, 2025. Tags: AI Engineering, Learning AI. An explanation of the Towards AI Partner Program, where referrals can earn books, courses, full access, or an affiliate tier. Questions: What is the Towards AI Partner Program? Who is this referral program for? What should partners avoid? Takeaways: The program is built around useful referrals: helping someone find the right AI learning path, not spamming links. Rewards start with the book, then courses, then full access and an affiliate tier for people who refer consistently.
- [Deep Research Agents: The Next Big Leap Beyond ChatGPT](https://www.louisbouchard.ai/deep-research-agents/) - Published Aug 26, 2025. Updated Aug 26, 2025. Tags: AI Agents, Large Language Models, AI News and Analysis. Deep research agents go beyond chat by browsing, planning, synthesizing, and producing cited reports, but they still need checks. Questions: What is a deep research agent? When are deep research agents useful? What should you verify in the output? Takeaways: Deep research agents are useful because they can spend minutes gathering sources and connecting them across a task. The output feels stronger than a normal chatbot answer, but it still depends on search quality, source choice, and synthesis. Audio: [original video audio](https://www.louisbouchard.ai/audio/deep-research-agents-7TEJe2ciWY4-v1.mp3), sourced from [OpenAI’s $200/Month Research Agent: Is It Worth It?](https://www.youtube.com/watch?v=7TEJe2ciWY4).
- [What actually changed with GPT-5](https://www.louisbouchard.ai/gpt-5/) - Published Aug 08, 2025. Updated Aug 08, 2025. Tags: Large Language Models. A practical read on GPT-5: what changed in reasoning, tools, cost, and when the full model beats smaller options. Questions: What actually changed with GPT-5? When should you use the full GPT-5 model? What is the model-selection mistake? Takeaways: The important GPT-5 question is not whether it is the best model overall. It is when the full model earns its cost. Mini and nano models can be the better engineering choice when latency, price, or task simplicity matter more. Audio: [original video audio](https://www.louisbouchard.ai/audio/gpt-5-w7lhj-BWa-c-v1.mp3), sourced from [GPT-5: hype or game-changer? Last fact is 🤯](https://www.youtube.com/watch?v=w7lhj-BWa-c).
- [2025: The Year of AI Agents](https://www.louisbouchard.ai/the-year-of-ai-agents/) - Published Jul 22, 2025. Updated Jul 22, 2025. Tags: AI Agents, Large Language Models. Why 2025 became the year of AI agents: longer context, cheaper tools, better models, and workflows that can finally act. Questions: Why is 2025 called the year of AI agents? What makes agents more practical now? What is the hype trap with AI agents? Takeaways: The agent hype has a technical base: longer context, better tool use, cheaper inference, and more reliable workflows. A model alone only chats. Agents become useful when tools, memory, retrieval, and actions are wrapped around it. Audio: [original video audio](https://www.louisbouchard.ai/audio/the-year-of-ai-agents-z0WsyHjmDf8-v1.mp3), sourced from [8 Big Changes Making AI Agents a Huge Deal!](https://www.youtube.com/watch?v=z0WsyHjmDf8).
- [LLMs Lie. Here’s How to Keep Them Honest](https://www.louisbouchard.ai/llms-lie/) - Published Jul 19, 2025. Updated Jul 19, 2025. Tags: AI Agents, Large Language Models, AI Engineering, Learning AI. A practical guide to keeping LLMs honest with grounding, verification, evals, and workflows that do not trust fluency alone. Questions: Why do LLMs lie? How do you keep an LLM honest? What is the dangerous mistake? Takeaways: LLMs do not need bad intent to mislead you. They can produce a polished answer without enough evidence. Honest systems need grounding, citations, checks, and evaluation loops around the model's output. Audio: [original video audio](https://www.louisbouchard.ai/audio/llms-lie-AJ8n0t7937U-v1.mp3), sourced from [Before You Build with AI, Watch THIS (Big Mistakes to Avoid)](https://www.youtube.com/watch?v=AJ8n0t7937U).
- [Prompt Engineering 101 for Business People](https://www.louisbouchard.ai/prompt-engineering-101/) - Published Jul 10, 2025. Updated Jul 10, 2025. Tags: Large Language Models, AI Engineering. A business-friendly introduction to prompt engineering: clear instructions, useful context, examples, and knowing when prompts are not enough. Questions: What is prompt engineering for business people? Do you need technical skills to write better prompts? What is the prompt engineering trap? Takeaways: Prompt engineering starts with clearly explaining the task, constraints, examples, and desired output. Good prompts can improve results, but they do not replace data, evaluation, workflow design, or domain judgment. Audio: [original video audio](https://www.louisbouchard.ai/audio/prompt-engineering-101-9f7CVmLnJnY-v1.mp3), sourced from [ChatGPT Hacks: 7 Tricks to Instantly Boost Your AI Results](https://www.youtube.com/watch?v=9f7CVmLnJnY).
- [How to Really Build on Top of LLMs](https://www.louisbouchard.ai/how-to-really-build-on-top-of-llms/) - Published Jul 04, 2025. Updated Jul 04, 2025. Tags: AI Agents, Large Language Models, RAG and Retrieval, AI Engineering. A practical session on building with LLMs, from prompts and RAG to evals, workflows, and the parts teams usually skip. Questions: What does it really mean to build on top of LLMs? Should teams start with prompts, RAG, or fine-tuning? What do teams usually skip? Takeaways: Building on LLMs is not only prompting. The hard parts are context, evaluation, reliability, cost, and product fit. RAG, fine-tuning, workflows, and agents solve different problems, so the use case has to come before the technique. Audio: [original video audio](https://www.louisbouchard.ai/audio/how-to-really-build-on-top-of-llms-kMiqtTbBbzI-v1.mp3), sourced from [Our Best Tips for Context Engineering (full training session) (from prompts to fine-tuning LLMs)](https://www.youtube.com/watch?v=kMiqtTbBbzI).
- [MCP: The USB Port for LLM Integrations](https://www.louisbouchard.ai/mcp/) - Published Jun 28, 2025. Updated Jun 28, 2025. Tags: AI Agents, Large Language Models. MCP acts like a standard connector layer for LLM tools, making integrations easier to share, inspect, and reuse. Questions: What is MCP for LLM integrations? Why do builders care about MCP? What should you be careful about with MCP? Takeaways: MCP is useful because LLM tools need a standard way to connect to files, apps, APIs, and services. The value is not only convenience. A shared protocol makes integrations easier to reuse and reason about. Audio: [original video audio](https://www.louisbouchard.ai/audio/mcp-3NUV8JcihCg-v1.mp3), sourced from [MCP Explained: Universal API for Prompts, Data, and Tools](https://www.youtube.com/watch?v=3NUV8JcihCg).
- [Our developer-focused guide to LLMs, free!](https://www.louisbouchard.ai/developer-focused-guide-to-llms/) - Published Jun 22, 2025. Updated Jun 22, 2025. Tags: Large Language Models, Learning AI. A free developer-focused LLM training on how models work, how to use them, and where their current limitations still show up. Questions: What does the free LLM training cover? Who is the developer-focused LLM guide for? What should developers avoid? Takeaways: Developers need more than prompt examples. They need to understand how LLMs behave and where they fail. The training covers model basics, interaction patterns, limitations, competitive advantages, and ways to reduce errors. Audio: [original video audio](https://www.louisbouchard.ai/audio/developer-focused-guide-to-llms-R5_udqy1L4s-v1.mp3), sourced from [AI Fundamentals Builders Need to Know to Fix LLM Limitations](https://www.youtube.com/watch?v=R5_udqy1L4s).
- [How Smart Are Reasoning Models in 2025?](https://www.louisbouchard.ai/reasoning-models/) - Published Jun 20, 2025. Updated Jun 20, 2025. Tags: Large Language Models. A practical explanation of reasoning models in 2025, why extra thinking time helps, and where the limits still are. Questions: How Smart Are Reasoning Models in 2025? What makes reasoning models different? Are reasoning models always smarter? Takeaways: Reasoning models improve some tasks by spending more computation at answer time instead of only scaling training. They can break problems into steps, but visible reasoning does not automatically mean the answer is correct. Audio: [original video audio](https://www.louisbouchard.ai/audio/reasoning-models-KOMJmG8eWE0-v1.mp3), sourced from [Reasoning Models Explained: The New Scaling Law Axis](https://www.youtube.com/watch?v=KOMJmG8eWE0).
- [LLM Weaknesses 101: What They Really Learn](https://www.louisbouchard.ai/llm-weaknesses-101/) - Published Jun 16, 2025. Updated Jun 16, 2025. Tags: Large Language Models, Learning AI. A practical explanation of what LLMs really learn, why they fail, and which weaknesses builders need to design around. Questions: What do LLMs really learn? Why do LLMs fail confidently? How should builders handle LLM weaknesses? Takeaways: LLMs learn patterns from data, which makes them powerful but also vulnerable to hallucination, bias, and shallow reasoning. A fluent answer can hide missing grounding, weak logic, or uncertainty the model does not know how to express. Audio: [original video audio](https://www.louisbouchard.ai/audio/llm-weaknesses-101-vmBLxEmlQIc-v1.mp3), sourced from [LLMs Don’t Think—They Predict. Here’s Why That Matters](https://www.youtube.com/watch?v=vmBLxEmlQIc).
- [Master LLMs in a Weekend](https://www.louisbouchard.ai/master-llms-in-a-weekend/) - Published Jun 10, 2025. Updated Jun 10, 2025. Tags: Large Language Models, Learning AI. A compact path for learning practical LLM development basics over a weekend, with enough structure to start building. Questions: Can you really learn LLMs in a weekend? What should you focus on first? What should weekend learners avoid? Takeaways: A weekend is enough to build a useful map of LLM development, but not enough to skip practice. Focus on prompts, APIs, context, retrieval, evals, and one small project instead of trying to learn every model.
- [How LLMs Learn](https://www.louisbouchard.ai/how-llms-learn/) - Published Jun 05, 2025. Updated Jun 05, 2025. Tags: Large Language Models, Learning AI. A beginner-friendly explanation of how LLMs learn from data, prediction, fine-tuning, and feedback without treating them as magic. Questions: How do LLMs learn? Does training make LLMs understand like humans? Why should builders understand training? Takeaways: LLMs learn by predicting patterns in huge amounts of text, not by understanding the world the way humans do. Fine-tuning and feedback can change behavior, but they do not turn the model into a grounded human reasoner. Audio: [original video audio](https://www.louisbouchard.ai/audio/how-llms-learn-7lHJidvROXs-v1.mp3), sourced from [A Brief Introduction to Large Language Models (LLMs) - AI for Business Professionals](https://www.youtube.com/watch?v=7lHJidvROXs).
- [Python Fundamentals + CS Concepts](https://www.louisbouchard.ai/python-fundamentals-cs-concepts/) - Published Apr 22, 2025. Updated Apr 22, 2025. Tags: Learning AI. A beginner guide to Python and computer science basics: syntax, variables, inputs, outputs, debugging, and how programs think. Questions: What Python fundamentals should beginners learn first? Do coding agents remove the need for fundamentals? What should beginners avoid while learning Python? Takeaways: Even with coding assistants, beginners still need the basic mental model of how programs run and fail. Python fundamentals are about syntax, data, control flow, functions, inputs, outputs, and debugging habits. Audio: [original video audio](https://www.louisbouchard.ai/audio/python-fundamentals-cs-concepts-_uRb5wlFhyw-v1.mp3), sourced from [Python Fundamentals + CS Concepts — A One‑Stop Starter Class](https://www.youtube.com/watch?v=_uRb5wlFhyw).
- [14 CEOs Break Down the Challenges in Quantum](https://www.louisbouchard.ai/14-ceos-break-down-the-challenges-in-quantum/) - Published Apr 08, 2025. Updated Apr 08, 2025. Tags: AI Engineering. Notes from Jensen Huang's GTC quantum panel on why quantum still feels hyped, where QPUs may help, and what has to work first. Questions: What was the main point from the GTC quantum panel? Where could quantum computing help first? What is the hype trap with quantum? Takeaways: Quantum still has a breakthrough-to-controversy problem, which makes practical use cases harder to judge from headlines alone. The near-term story is less about replacing classical computers and more about finding narrow places where QPUs can help. Audio: [original video audio](https://www.louisbouchard.ai/audio/14-ceos-break-down-the-challenges-in-quantum-wlfMDm6qGOE-v1.mp3), sourced from [The Future of Quantum from 14 CEOs! Insights from GTC 2025](https://www.youtube.com/watch?v=wlfMDm6qGOE).
- [An Introduction to Colab and Jupyter for Beginners](https://www.louisbouchard.ai/colab-vs-jupyter/) - Published Mar 31, 2025. Updated Mar 31, 2025. Tags: Learning AI, AI News and Analysis. A beginner-friendly comparison of Google Colab and Jupyter notebooks, with when to use each for Python and AI work. Questions: Should beginners start with Colab or Jupyter? When is Jupyter the better choice? What should beginners avoid with notebooks? Takeaways: Both Colab and Jupyter let you run Python in cells, mix code with notes, and see outputs as you learn. Colab is easier to start because it runs in the browser and gives you hosted compute without local setup. Audio: [original video audio](https://www.louisbouchard.ai/audio/colab-vs-jupyter-rkg9qIh6_0c-v1.mp3), sourced from [The BIGGEST Differences Between Jupyter and Colab You Need to Know!](https://www.youtube.com/watch?v=rkg9qIh6_0c).
- [Open-Sora 2.0 Explained: Architecture, Training, and Why It Matters](https://www.louisbouchard.ai/open-sora-2/) - Published Mar 28, 2025. Updated Mar 28, 2025. Tags: AI Engineering, Generative AI. A practical walkthrough of Open-Sora 2.0, its video generation architecture, training approach, and why open video models matter. Questions: What is Open-Sora 2.0? Why is video generation harder than images? What should you watch in video model demos? Takeaways: Open video models matter because they let more builders inspect, test, and improve the pieces behind generation. Video generation is harder than image generation because motion, identity, physics, and temporal consistency all have to hold together. Audio: [original video audio](https://www.louisbouchard.ai/audio/open-sora-2-gMdvyGVICfA-v1.mp3), sourced from [How Open-Sora 2.0 Built Sora-Level Video AI for $200K (Full Breakdown)](https://www.youtube.com/watch?v=gMdvyGVICfA).
- [Inside ChatGPT: How AI Understands and Generates Language](https://www.louisbouchard.ai/inside-chatgpt/) - Published Mar 22, 2025. Updated Mar 22, 2025. Tags: Large Language Models, Generative AI, Learning AI. A beginner-friendly explanation of how ChatGPT turns text into tokens, predicts responses, and generates language that feels coherent. Questions: How does ChatGPT understand language? Why can ChatGPT sound so coherent? What is the main limitation to remember? Takeaways: ChatGPT does not read text the way humans do. It works with tokens and predicts likely continuations. The impressive part is not one prediction, but how many small predictions combine into a useful answer. Audio: [original video audio](https://www.louisbouchard.ai/audio/inside-chatgpt-PcZBqyxRfXw-v1.mp3), sourced from [AI Explained For Complete Beginners - No Math! Towards AI's Python Primer for Generative AI](https://www.youtube.com/watch?v=PcZBqyxRfXw).
- [How FlashMLA Cuts KV Cache Memory to 6.7%](https://www.louisbouchard.ai/flashmla/) - Published Mar 19, 2025. Updated Mar 19, 2025. Tags: AI Agents, Large Language Models. A plain-English look at FlashMLA, DeepSeek's KV cache memory trick, and why efficient attention matters for large models. Questions: What problem does FlashMLA solve? Why does KV cache memory matter? What should builders watch with efficiency claims? Takeaways: The KV cache stores attention information so a model does not recompute everything for every new token. FlashMLA matters because memory, not only raw compute, becomes a bottleneck when serving long or large model requests. Audio: [original video audio](https://www.louisbouchard.ai/audio/flashmla-PTNW3SLP8W8-v1.mp3), sourced from [DeepSeek's FlashMLA Explained](https://www.youtube.com/watch?v=PTNW3SLP8W8).
- [OpenAI's NEW Fine-Tuning Method Changes EVERYTHING](https://www.louisbouchard.ai/rft/) - Published Mar 16, 2025. Updated Mar 16, 2025. Tags: Large Language Models, AI News and Analysis. A practical explanation of reinforcement fine-tuning, how it differs from supervised fine-tuning, and when rewards help. Questions: What is reinforcement fine-tuning? How is RFT different from supervised fine-tuning? What can go wrong with RFT? Takeaways: Supervised fine-tuning teaches a model to imitate examples. Reinforcement fine-tuning teaches it through rewards. RFT can help when correct behavior is easier to judge than to demonstrate with a huge dataset. Audio: [original video audio](https://www.louisbouchard.ai/audio/rft-i40tCb7bkmg-v1.mp3), sourced from [What is Reinforcement Fine-Tuning (RFT) - Supervised vs. RL LLM Re-training](https://www.youtube.com/watch?v=i40tCb7bkmg).
- [The Era of "ChatGPT Programming"](https://www.louisbouchard.ai/the-era-of-chatgpt-programming/) - Published Mar 06, 2025. Updated Mar 06, 2025. Tags: Large Language Models. An introduction to programming with ChatGPT support, focused on giving computers clear instructions and understanding the code. Questions: What is ChatGPT programming? Can beginners learn programming with ChatGPT? What should beginners be careful about? Takeaways: Programming is giving precise instructions to a computer, even if ChatGPT helps draft the code. AI can speed up learning, but beginners still need to understand steps, errors, and how a program flows. Audio: [original video audio](https://www.louisbouchard.ai/audio/the-era-of-chatgpt-programming-F1COwAB_g_E-v1.mp3), sourced from [How LLMs Can be Used as Coding Assistants (Python for AI Beginners Course by Towards AI)](https://www.youtube.com/watch?v=F1COwAB_g_E).
- [From Zero to LLM Hero: Announcing Our Python Primer for Generative AI](https://www.louisbouchard.ai/from-zero-to-llm-hero/) - Published Feb 27, 2025. Updated Feb 27, 2025. Tags: Large Language Models, Generative AI, Learning AI. An announcement for a Python primer built to help beginners move from shaky coding basics toward practical generative AI work. Questions: Who is From Zero to LLM Hero for? Why start with Python before LLM projects? What should Python beginners avoid? Takeaways: Many people want to build with LLMs but get blocked by basic Python, debugging, and project structure. The primer is meant to remove that early friction before learners move into deeper generative AI development. Audio: [original video audio](https://www.louisbouchard.ai/audio/from-zero-to-llm-hero-_hOpfENqPPA-v1.mp3), sourced from [A Python Primer for Generative AI | From Coding Novice to Building with LLMs](https://www.youtube.com/watch?v=_hOpfENqPPA).
- [Will AI Take Your Job?](https://www.louisbouchard.ai/will-ai-take-your-job/) - Published Feb 11, 2025. Updated Feb 11, 2025. Tags: Large Language Models. A practical look at how LLMs affect jobs, which tasks are easier to automate, and why adaptation beats panic. Questions: Will AI Take Your Job? Will AI take your job? Which tasks are most exposed to LLMs? Takeaways: LLMs affect tasks before they affect entire jobs, especially repetitive text, analysis, and recognition-heavy work. Many roles will shift toward human judgment, domain context, communication, and checking AI-assisted output. Audio: [original video audio](https://www.louisbouchard.ai/audio/will-ai-take-your-job-1iHJGTlyDqc-v1.mp3), sourced from [How LLMs Are Changing Every Job (And What You Can Do About It)](https://www.youtube.com/watch?v=1iHJGTlyDqc).
- [Staying Ahead of AI Developments in 2025](https://www.louisbouchard.ai/ai-developments-in-2025/) - Published Feb 10, 2025. Updated Feb 10, 2025. Tags: Learning AI, AI News and Analysis. A practical guide to tracking AI in 2025 without chasing every release, by narrowing sources around your work and goals. Questions: How should you keep up with AI in 2025? Which AI updates are worth following? What is the trap in following AI news? Takeaways: No one can follow every AI update, so the useful skill is filtering by your project, job, or learning goal. Good AI tracking starts with a small set of trusted newsletters, communities, researchers, and product sources.
- [The Best 'LLM Developer' Courses in 2025](https://www.louisbouchard.ai/the-best-llm-developer-courses-in-2025/) - Published Feb 09, 2025. Updated Feb 09, 2025. Tags: Large Language Models, Learning AI. A practical guide to choosing LLM developer courses in 2025 based on projects, depth, and the skills companies actually need. Questions: What should an LLM developer course teach? How should beginners choose a course? What is the course-selection trap? Takeaways: The best LLM course is not the one with the loudest promise. It is the one that helps you build and evaluate real systems. Look for coverage of prompting, RAG, fine-tuning, agents, deployment, and the judgment needed to choose between them.
- [LLM Developers vs Software Developers vs ML Engineers: Key Differences](https://www.louisbouchard.ai/llm-developers/) - Published Feb 05, 2025. Updated Feb 11, 2025. Tags: Large Language Models, AI Engineering. A comparison of LLM developers, software developers, and ML engineers, focused on the skills each role actually needs. Questions: How is an LLM developer different from a software developer? Do LLM developers need machine learning skills? What should learners focus on first? Takeaways: LLM development overlaps with software engineering, but adds model behavior, context, retrieval, and evaluation work. ML engineering is still relevant, but many LLM products rely more on system design than training a model from scratch. Audio: [original video audio](https://www.louisbouchard.ai/audio/llm-developers-UbLJuxSqEN0-v1.mp3), sourced from [The future of software development: LLM Developers](https://www.youtube.com/watch?v=UbLJuxSqEN0).
- [Polish Your Writing With AI](https://www.louisbouchard.ai/edition-avec-ia/) - Published Feb 03, 2025. Updated Feb 03, 2025. Tags: Large Language Models, Generative AI. A practical workflow for using ChatGPT or Claude to revise a lesson, add explanations, produce a conclusion, and verify sources. Questions: How can you use AI to revise a first draft? Why ask for suggestions before a complete rewrite? Why assign IDs to the suggestions? Takeaways: Ask the LLM for targeted suggestions before giving it a vague rewriting task. Use IDs and a precise output format to choose which changes the model should apply.
- [Agents or Workflows?](https://www.louisbouchard.ai/agents-vs-workflows/) - Published Feb 02, 2025. Updated Feb 02, 2025. Tags: AI Agents, Large Language Models. A practical way to separate real agents from LLM calls and workflows, so teams choose the right amount of autonomy. Questions: Agents or Workflows? What makes something a real AI agent? When is a workflow better than an agent? Takeaways: A raw LLM API call is not an agent. It replies, but it does not act independently or choose a path. Tool use alone still does not guarantee agency if the system is mostly following hard-coded steps. Audio: [original video audio](https://www.louisbouchard.ai/audio/agents-vs-workflows-kQxr-uOxw2o-v1.mp3), sourced from [Real Agents vs. Workflows: The Truth Behind AI 'Agents'](https://www.youtube.com/watch?v=kQxr-uOxw2o).
- [CAG vs RAG: Which One is Right for You?](https://www.louisbouchard.ai/cag-vs-rag/) - Published Jan 29, 2025. Updated Jan 29, 2025. Tags: Large Language Models, RAG and Retrieval, AI Engineering. A practical comparison of CAG and RAG for LLM context, including when cached context helps and when retrieval still wins. Questions: CAG vs RAG: Which One is Right for You? What is the difference between CAG and RAG? When should you use CAG instead of RAG? Takeaways: RAG searches external sources at query time, while CAG tries to preload the needed context into the model's cache. CAG can simplify the system when the knowledge base is small, stable, and fits inside the context you can afford. Audio: [original video audio](https://www.louisbouchard.ai/audio/cag-vs-rag-Z-rEACwLIqE-v1.mp3), sourced from [What is Cache Augmented Generation (CAG) - CAG vs RAG](https://www.youtube.com/watch?v=Z-rEACwLIqE).
- [Why LLMs Are the Future of Work](https://www.louisbouchard.ai/why-llms-are-the-future-of-work/) - Published Jan 27, 2025. Updated Jan 27, 2025. Tags: Large Language Models. A practical look at how LLMs change work, which tasks they affect first, and which skills help people stay competitive. Questions: Why are LLMs changing work? Will LLMs replace every job? How can workers stay competitive? Takeaways: LLMs are most useful where work involves text, analysis, drafting, coding assistance, summarization, or repeated knowledge tasks. They are more likely to augment many skilled jobs before fully replacing them, but the task mix will change. Audio: [original video audio](https://www.louisbouchard.ai/audio/why-llms-are-the-future-of-work-Fv0jRqYGFLg-v1.mp3), sourced from [7 Reasons Why Learning LLMs is No Longer Optional](https://www.youtube.com/watch?v=Fv0jRqYGFLg).
- [Optimize Your Writing Process With AI](https://www.louisbouchard.ai/ecriture-avec-ia/) - Published Jan 26, 2025. Updated Jan 26, 2025. Tags: Large Language Models, Generative AI. How to use LLMs to write better lessons without losing your ideas, judgment, sources, or personal style. Questions: Why shouldn't you ask ChatGPT to write the entire lesson? What context should you provide to an LLM for writing? Why use at least three sources? Takeaways: Start with your own plan and ideas before involving an LLM. Give the model multiple reliable sources and precise instructions.
- [Research With AI Tools](https://www.louisbouchard.ai/rechercher-avec-des-outils-dia/) - Published Jan 22, 2025. Updated Jan 23, 2025. Tags: Large Language Models, RAG and Retrieval, AI Engineering. Use Perplexity, ChatGPT Search, Ai2 OpenScholar, and RAG to research faster, verify sources, and work with local documents. Questions: Why use an AI tool for research? What is the difference between Perplexity and Ai2 OpenScholar? Do citations from an AI tool eliminate hallucinations? Takeaways: AI research tools combine an LLM with web sources retrieved in real time. You should always open and verify citations before reusing a claim.
- [Introducing Our 8-Hour Generative AI Primer](https://www.louisbouchard.ai/introducing-our-8-hour-generative-ai-primer/) - Published Jan 16, 2025. Updated Jan 16, 2025. Tags: Large Language Models, Generative AI, Learning AI. An introduction to an 8-hour generative AI primer built to help learners understand models, prompting, and practical use cases. Questions: Who is the generative AI primer for? What should a beginner learn first? What should generative AI beginners avoid? Takeaways: A good generative AI primer should connect the basics to real use cases instead of only listing tools. Learners need enough model intuition to understand prompting, limitations, and when outputs need checking. Audio: [original video audio](https://www.louisbouchard.ai/audio/introducing-our-8-hour-generative-ai-primer-6rBtV47JWog-v1.mp3), sourced from [8-hour Generative AI Primer for Software Professionals](https://www.youtube.com/watch?v=6rBtV47JWog).
- [Best Practices for Building and Deploying Scalable APIs in 2025](https://www.louisbouchard.ai/deploying-scalable-apis/) - Published Jan 11, 2025. Updated Jan 23, 2025. Tags: AI Engineering, Learning AI. A practical API deployment guide for AI engineers who need models, data, and products to talk reliably in production. Questions: Why do AI engineers need APIs? What matters when deploying an AI API? What is the common deployment mistake? Takeaways: An API is the bridge between your AI system and the product, user, database, or service that needs it. Deployment is not only hosting a model. It also includes reliability, latency, scaling, monitoring, and clear interfaces. Audio: [original video audio](https://www.louisbouchard.ai/audio/deploying-scalable-apis-5atR70lV1fs-v1.mp3), sourced from [APIs 101: From Concept to Deployment for AI Engineers](https://www.youtube.com/watch?v=5atR70lV1fs).
- [Lessons from Minitron](https://www.louisbouchard.ai/lessons-from-minitron/) - Published Jan 08, 2025. Updated Jan 23, 2025. Tags: Large Language Models. Lessons from Minitron on building smaller language models, where efficiency comes from, and why model size is not the full story. Questions: What is the main lesson from Minitron? When should builders consider a smaller model? What is the trap with efficient models? Takeaways: Smaller models can be useful when the task, data, and deployment constraints do not justify a larger model. Efficiency gains usually come from tradeoffs, not magic. You still need to test quality on your own workload. Audio: [original video audio](https://www.louisbouchard.ai/audio/lessons-from-minitron-eJs-8IDHJ3w-v1.mp3), sourced from [Pruning and Distillation Best Practices: The Minitron Approach Explained](https://www.youtube.com/watch?v=eJs-8IDHJ3w).
- [When NOT to Use Large Language Models](https://www.louisbouchard.ai/when-not-to-use-large-language-models/) - Published Jan 03, 2025. Updated Jan 23, 2025. Tags: Large Language Models, Learning AI. A practical guide to when LLMs are overkill, when simpler models work better, and how to decide before spending the compute. Questions: When should you avoid using LLMs? When do LLMs make sense? What is the expensive LLM mistake? Takeaways: LLMs are useful for language-heavy, open-ended, or creative tasks, but they are not the right tool for everything. If the task is narrow, structured, cheap to solve, or needs deterministic behavior, a simpler model may be better. Audio: [original video audio](https://www.louisbouchard.ai/audio/when-not-to-use-large-language-models-_gG01yyqpaQ-v1.mp3), sourced from [To use a LLM or to not use it?](https://www.youtube.com/watch?v=_gG01yyqpaQ).
- [Optimizing RAG: Should you Fine-Tune?](https://www.louisbouchard.ai/optimizing-rag-should-you-fine-tune/) - Published Dec 21, 2024. Updated Jan 23, 2025. Tags: Large Language Models, RAG and Retrieval, AI Engineering, Generative AI. Optimizing Large Language Models for Retrieval-Augmented Generation. How do you deploy these models in a way that’s even better? Questions: Optimizing RAG: Should you Fine-Tune? Should every RAG system fine-tune its language model? When can fine-tuning improve RAG? Takeaways: How do you deploy these models in a way that’s even better? This is especially useful if you’ve trained a model in one environment but need to deploy or fine-tune it in another. Audio: [original video audio](https://www.louisbouchard.ai/audio/optimizing-rag-should-you-fine-tune-CWTSImmqcvQ-v1.mp3), sourced from [Fine-Tuning LLMs for RAG: Boost Model Performance and Accuracy](https://www.youtube.com/watch?v=CWTSImmqcvQ).
- [What Are Vector Databases?](https://www.louisbouchard.ai/what-are-vector-databases/) - Published Dec 12, 2024. Updated Jan 23, 2025. Tags: RAG and Retrieval. Building AI-Powered Search and Recommendations. As your data grows, you’ll need a database that can scale with it. Questions: What Are Vector Databases? What is a vector database? Why are vectors useful for search? Takeaways: As your data grows, you’ll need a database that can scale with it. You need constant, high-speed data, which requires a lot of hardware, while your friend doesn’t need as much. Audio: [original video audio](https://www.louisbouchard.ai/audio/what-are-vector-databases-TprMWe54kZE-v1.mp3), sourced from [Vector Databases Explained (vs traditional databases & alternatives)](https://www.youtube.com/watch?v=TprMWe54kZE).
- [Advanced Search Techniques: From Keywords to Graphs](https://www.louisbouchard.ai/advanced-search-techniques/) - Published Dec 03, 2024. Updated Jan 23, 2025. Tags: Large Language Models, RAG and Retrieval. Build a Smarter RAG System. If you need to understand meaning and context, go for embedding search. Questions: When should I use keyword search in a RAG system? What does embedding search add? Why combine keyword and embedding search? Takeaways: If you need to understand meaning and context, go for embedding search. Instead of simply penalizing the document, BM42 considers other factors, like the overall context or related terms, to better assess the document’s relevance. Audio: [original video audio](https://www.louisbouchard.ai/audio/advanced-search-techniques-tSyDodhNE-A-v1.mp3), sourced from [Techniques for Optimizing Search in RAG Systems](https://www.youtube.com/watch?v=tSyDodhNE-A).
- [How to Improve Your RAG Pipeline with Fine-Tuned Embedding Models](https://www.louisbouchard.ai/fine-tuned-embedding-models/) - Published Nov 29, 2024. Updated Nov 29, 2024. Tags: Large Language Models, RAG and Retrieval, AI Engineering. Improve a RAG pipeline with fine-tuned embeddings, better retrieval checks, and a clearer view of when the extra work is worth it. Questions: What does an embedding model do in RAG? When can fine-tuning embeddings improve retrieval? Do all RAG systems need fine-tuned embeddings? Takeaways: The embedding model creates numerical representations of your data; the vector database stores and retrieves those representations. Fine-tuning reshapes the embedding space so domain-specific queries and documents are more likely to land near each other. Audio: [original video audio](https://www.louisbouchard.ai/audio/fine-tuned-embedding-models-YE6-dS-80KQ-v1.mp3), sourced from [Boost RAG Performances with Fine-Tuned Embedding Models](https://www.youtube.com/watch?v=YE6-dS-80KQ).
- [Long Context Models Explained: Do We Still Need RAG?](https://www.louisbouchard.ai/long-context-vs-rag/) - Published Nov 26, 2024. Updated Jan 23, 2025. Tags: Large Language Models, RAG and Retrieval, AI Engineering, Generative AI. A practical comparison of long-context models and RAG across data volume, retrieval quality, latency, cost, and the workflows where each approach fits. Questions: Long Context Models Explained: Do We Still Need RAG? What is a long-context language model? Why use RAG when the context window is already large? Takeaways: Long-context models can simplify one-off work with a few documents because the full input may fit in one prompt without a retrieval pipeline. RAG remains useful for large or changing datasets because it retrieves a smaller, more relevant context instead of sending the whole collection on every request. Audio: [original video audio](https://www.louisbouchard.ai/audio/long-context-vs-rag-qN3vhWlzd4A-v1.mp3), sourced from [RAG vs. Long Context Models: Is Retrieval-Augmented Generation Dead?](https://www.youtube.com/watch?v=qN3vhWlzd4A).
- [Don’t Fall for the Hype: How to Identify Truly Useful AI](https://www.louisbouchard.ai/hype-or-not/) - Published Nov 22, 2024. Updated Jan 23, 2025. Tags: Large Language Models, AI Engineering. What Makes AI Truly Useful? This way, since it’s built into the tool you already use, you don’t need to be a master prompt engineer or explain technicalities. Questions: What makes an AI feature genuinely useful? Why does integration matter more than a flashy demo? How does connected workspace AI save time? Takeaways: This way, since it’s built into the tool you already use, you don’t need to be a master prompt engineer or explain technicalities. Useful AI tackles specific, often tedious tasks, while AI, for the sake of AI, tends to over-engineer simple processes. Audio: [original video audio](https://www.louisbouchard.ai/audio/hype-or-not-EHuxiQoqD1k-v1.mp3), sourced from [When is AI Integration REALLY Worth the Hype?](https://www.youtube.com/watch?v=EHuxiQoqD1k).
- [How to Effectively Evaluate Retrieval-Augmented Generation (RAG) Systems](https://www.louisbouchard.ai/rag-evals/) - Published Nov 21, 2024. Updated Jan 23, 2025. Tags: RAG and Retrieval, AI Engineering, Generative AI. Evaluate RAG systems with retrieval and answer-quality checks that show whether your pipeline is actually improving. Questions: Why evaluate retrieval separately from generation? What should a RAG evaluation dataset contain? Which retrieval qualities should be measured? Takeaways: Let’s explain the key evaluation metrics and methods we’ve found useful while developing RAG systems at Towards AI. For example, if you are building a legal research tool with RAG, it will need to be contextually appropriate and up-to-date. Audio: [original video audio](https://www.louisbouchard.ai/audio/rag-evals-cRz0BWkuwHg-v1.mp3), sourced from [Key Metrics and Evaluation Methods for RAG](https://www.youtube.com/watch?v=cRz0BWkuwHg).
- [Indexing Methods for Vector Retrieval](https://www.louisbouchard.ai/indexing-methods/) - Published Nov 18, 2024. Updated Jan 23, 2025. Tags: Large Language Models, RAG and Retrieval, AI Engineering, Learning AI. Learn how HNSW, LSH, product quantization, and other vector indexing methods make RAG retrieval faster, more accurate, and scalable. Questions: Why does vector retrieval need an index? What is exact nearest-neighbor search? What does approximate nearest-neighbor search trade? Takeaways: As we’ve discussed in our book and course, retrieving documents or useful text is necessary for all LLM-based products. You need to give additional information to your LLM if you want it to be really useful. Audio: [original video audio](https://www.louisbouchard.ai/audio/indexing-methods-NytKzh8avhw-v1.mp3), sourced from [What is Indexing? Indexing Methods for Vector Retrieval](https://www.youtube.com/watch?v=NytKzh8avhw).
- [Releasing our 90+ lesson practical LLM Developer course!](https://www.louisbouchard.ai/rag-course-tai-academy/) - Published Nov 14, 2024. Updated Nov 14, 2024. Tags: Large Language Models, RAG and Retrieval, Learning AI. I am super excited to finally announce that we ( Towards AI ) released our first independent industry-focussed course: From Beginner to Advanced LLM Developer. Questions: What is the From Beginner to Advanced LLM Developer course? Who created the course? What makes the course industry-focused? Takeaways: Put a dozen experts (frustrated ex-PhDs, graduates and industry) and a year of dedicated work. Here is our platform introduction video to give you more context on our goal with the Towards AI Academy: So we’ve gathered everything we worked on building products. Audio: [original video audio](https://www.louisbouchard.ai/audio/rag-course-tai-academy-tlP1w_nligI-v1.mp3), sourced from [Master LLMs and Get Industry-ready - The Towards AI Academy](https://www.youtube.com/watch?v=tlP1w_nligI).
- [Find the Best AI Model for Your Specific Task](https://www.louisbouchard.ai/evaluating-llms/) - Published Nov 13, 2024. Updated Nov 13, 2024. Tags: Large Language Models, AI Engineering. No-code Custom LLM Evaluation Demo. You just create your benchmark with questions and answers the LLM should provide. Questions: Why evaluate LLMs on a custom benchmark? What should a task-specific benchmark contain? Can an LLM grade another model's output? Takeaways: You just create your benchmark with questions and answers the LLM should provide. You can finally get an overall idea of what model performs best for your task and dive in more to see why it generates better results. Audio: [original video audio](https://www.louisbouchard.ai/audio/evaluating-llms-lX1oLcgkZXg-v1.mp3), sourced from [Are You Choosing the WRONG LLM for Your Project?](https://www.youtube.com/watch?v=lX1oLcgkZXg).
- [Building a Multi-Agent System (No-Code)](https://www.louisbouchard.ai/multi-agent-system/) - Published Nov 11, 2024. Updated Jan 23, 2025. Tags: AI Agents, AI Engineering. No-code Agentic AI Demo. They then allow you to deploy the system and create an API out of it, all without any code knowledge. Questions: What is a no-code multi-agent system? When are specialized agents useful? Can different agents use different models? Takeaways: They then allow you to deploy the system and create an API out of it, all without any code knowledge. ReACT, developed as a method that combines reasoning and acting, is incredibly effective for multi-agent systems that require both logical thinking and practical actions. Audio: [original video audio](https://www.louisbouchard.ai/audio/multi-agent-system-mhjIRis4SmI-v1.mp3), sourced from [Master Multi-Agent Systems Like a PRO with AGENTIC AI](https://www.youtube.com/watch?v=mhjIRis4SmI).
- [Top RAG Techniques You Should Know (Wang et al., 2024)](https://www.louisbouchard.ai/top-rag-techniques/) - Published Sep 13, 2024. Updated Sep 13, 2024. Tags: Large Language Models, RAG and Retrieval, Generative AI. Explore the RAG techniques tested by Wang et al. (2024), from query rewriting and reranking to fine-tuning the embedding model and generator. Questions: Does every user question require retrieval? What is query routing in RAG? Why rerank retrieved passages? Takeaways: Fine-tuning the Generator Lastly, should you fine-tune the LLM you are using for generation? Not all queries are created equal, some don’t even need retrieval because the large language model already knows the answer. Audio: [original video audio](https://www.louisbouchard.ai/audio/top-rag-techniques-QyaMmSXxH2w-v1.mp3), sourced from [The Best RAG Stack Components to date (fully open-source!)](https://www.youtube.com/watch?v=QyaMmSXxH2w).
- [OpenAI's o1 Model: The Future of Reasoning AI? What Sets It Apart](https://www.louisbouchard.ai/openai-o1/) - Published Sep 13, 2024. Updated Jan 23, 2025. Tags: Large Language Models, AI News and Analysis. Unfortunately, as with some of us, this process of thinking before answering makes o1 much, much slower than previous models. Questions: What makes OpenAI's o1 different from GPT-4o? Why is o1 slower? When is o1 a useful choice? Takeaways: Indeed, it’s slower than models like GPT-4o because it takes time to think through problems, but if you need a model that excels at solving complex tasks. If you're working with complicated problems or need a model that's reliable across different languages, o1 is definitely worth the extra wait time. Audio: [original video audio](https://www.louisbouchard.ai/audio/openai-o1-ICzPJ8zJRhY-v1.mp3), sourced from [Is OpenAI's latest o1 model that good?](https://www.youtube.com/watch?v=ICzPJ8zJRhY).
- [GAIA-Graph, Also Known as GraphRAG, Explained](https://www.louisbouchard.ai/gaia-graphe-alias-graphrag/) - Published Aug 22, 2024. Updated Aug 22, 2024. Tags: RAG, GraphRAG, Large Language Models. Learn how GraphRAG adds knowledge graphs to RAG, when that extra complexity helps, and when a simpler retrieval approach is the better choice. Questions: What is GAIA-Graph or GraphRAG? When is GraphRAG more useful than traditional RAG? What kind of data works well with GraphRAG? Takeaways: GAIA-Graph enriches RAG with a knowledge graph that explicitly represents entities and their relationships. It is most useful for interconnected data and complex queries that require traversing several relationships.
- [Large Language Models: More Than Autocomplete](https://www.louisbouchard.ai/grands-modeles-de-langues-completeurs/) - Published Aug 14, 2024. Updated Jan 23, 2025. Tags: Large Language Models, Learning AI. Why large language models do more than complete text, and how context, attention, and chain-of-thought prompting change their answers. Questions: Is an LLM only an advanced autocomplete system? Where does a word's meaning come from for a language model? What does attention do in an LLM? Takeaways: A word gets its meaning from the sentence context and the patterns learned from training data. Attention mechanisms help LLMs connect the relevant words in a context.
- [Personalized Marketing with AI](https://www.louisbouchard.ai/marketing-with-ai/) - Published Aug 14, 2024. Updated Aug 14, 2024. Tags: Large Language Models. The Real AI Tools Revolutionizing Modern Marketing Beyond ChatGPT. We also need to be careful with the data collected from our users and use it ethically. Questions: How can AI personalize marketing? What can AI add beyond a generic chatbot? How can sentiment analysis help a brand? Takeaways: We also need to be careful with the data collected from our users and use it ethically. From simple scripts to smart AI-driven systems using NLP and machine learning, they can now understand past interactions and personal data to create really personalized. Audio: [original video audio](https://www.louisbouchard.ai/audio/marketing-with-ai-EnT28ui1ZX0-v1.mp3), sourced from [AI tools for marketers (beyond ChatGPT...)](https://www.youtube.com/watch?v=EnT28ui1ZX0).
- [The Myth of “Advanced Prompts”](https://www.louisbouchard.ai/le-mythe-des-prompts-avances/) - Published Jul 31, 2024. Updated Jul 31, 2024. Tags: Prompting, Large Language Models, AI Engineering. So-called advanced prompting techniques are mostly good communication. Here are the few methods that really matter and how to structure an LLM’s output. Questions: What makes a prompt effective? Are advanced prompts really advanced? What is the difference between zero-shot and few-shot prompting? Takeaways: Effective prompting starts with clear, concise communication and relevant examples when they help. Zero-shot, few-shot, chain-of-thought, chaining, and RAG are technical names for fairly simple ideas.
- [Why Is Llama 3.1 So Important?](https://www.louisbouchard.ai/pourquoi-llama-31-est-il-si-important/) - Published Jul 25, 2024. Updated Jul 25, 2024. Tags: Llama 3.1, Open Source Models, Large Language Models. Eleven essential questions about Llama 3.1 covering performance, cost, infrastructure, tooling, and why it matters to businesses. Questions: Why Is Llama 3.1 So Important? Why is Llama 3.1 important? What are the main advantages of an open source model? Takeaways: Llama 3.1 makes a frontier 405-billion-parameter model available with open weights and a license that permits distillation. Businesses gain control, customization, and infrastructure choice, but must plan for the necessary skills and GPUs.
- [How Nvidia trained Nemotron](https://www.louisbouchard.ai/nemotron-340b/) - Published Jul 08, 2024. Updated Jul 08, 2024. Tags: Large Language Models, AI News and Analysis. Training LLMs with Synthetic Data. Okay, so that’s cool and all; we’ve got a way to create better models with little manual data improvement work. Questions: How did NVIDIA use synthetic data for Nemotron? Why generate synthetic data? What makes synthetic data useful rather than repetitive? Takeaways: Okay, so that’s cool and all; we’ve got a way to create better models with little manual data improvement work. The secret is the enormous amount of high-quality data these models need.
- [LLM Evals](https://www.louisbouchard.ai/llm-evals/) - Published Jun 24, 2024. Updated Jan 23, 2025. Tags: Large Language Models, AI Engineering. What, why, when and how…. We then check the actual accuracy in each group: For the predictions the model is 70% confident about. Questions: Why do teams evaluate LLMs? What does an LLM evaluation dataset need? Who can grade model responses? Takeaways: We then check the actual accuracy in each group: For the predictions the model is 70% confident about, we need to confirm if about 70% of them are actually correct. This means we need a dataset with questions and a way to compare responses (our metrics).
- [How to Build a Multimodal LLM like GPT-4o?](https://www.louisbouchard.ai/chameleon/) - Published Jun 14, 2024. Updated Jun 14, 2024. Tags: Large Language Models, Computer Vision, AI News and Analysis. A clear look at Chameleon, the multimodal LLM idea behind GPT-4o-style systems, and what changes for text, image, and video models. Questions: How to Build a Multimodal LLM like GPT-4o? What is a multimodal language model? How is Chameleon different from a text-only LLM? Takeaways: Now that we’ve looked into the “how” of building a multimodal model, where exactly is this useful and differs from other models, like GPT-4 or Llama? When a model works with just one type, like GPT-4 for text, it’s unimodal.
- [The Hidden Dangers of AI in Coding: What You Need to Know](https://www.louisbouchard.ai/genai-coding-risks/) - Published May 30, 2024. Updated May 30, 2024. Tags: Generative AI. If you are coding with ChatGPT or Copilot, you may be creating some terrible security leaks! Questions: Why are AI coding tools useful? What security risks can generated code introduce? Can developers paste proprietary code into any chatbot? Takeaways: It’s especially useful and important for larger companies or when working with multiple developers if you are trying to build a reliable and valuable codebase. Just ask any current undergrad software engineer about Copilot or ChatGPT, and you’ll see.
- [The AI Engineer's toolkit](https://www.louisbouchard.ai/book-towards-ai/) - Published May 28, 2024. Updated Nov 11, 2024. Tags: Large Language Models, RAG and Retrieval, AI Engineering. Our first book: Building LLMs for Production: Enhancing LLM Abilities and Reliability with Prompting, Fine-Tuning, and RAG. Questions: What does The AI Engineer's Toolkit cover? Who is the book written for? Why does the book include RAG? Takeaways: This book provides practical insights and real-world applications of, inter alia, RAG systems and prompt engineering. It starts from explaining what an LLM is in simpler terms, and takes you through a brief history of time in NLP to the most current state of technology in AI.
- [How LLMs Know When to Stop Talking?](https://www.louisbouchard.ai/how-llms-know-when-to-stop/) - Published May 09, 2024. Updated May 09, 2024. Tags: Large Language Models. Learn how LLMs like GPT-4 use EOS tokens and maximum token limits to decide when a response is complete. Questions: How LLMs Know When to Stop Talking? How does an LLM decide to stop generating? What is an EOS token? Takeaways: For instance, when you ask a simple question like ‘What is the capital of France?’, the model generates a response and then places the EOS token right after ‘Paris. Two scenarios can make the model stop generating: ‘EOS tokens’ ( ) and ‘Maximum Token Lengths.’ We will learn about them.
- [Google's Infini-Attention](https://www.louisbouchard.ai/infini-attention/) - Published Apr 18, 2024. Updated Apr 18, 2024. Tags: Large Language Models, AI News and Analysis. How Google's Infini-attention combines local attention with compressive memory to process long sequences without recalculating the full history. Questions: What problem does Infini-attention address? How does Infini-attention handle recent context? How does it retain older information? Takeaways: A larger context window does not guarantee better answers. Models can lose important details in long inputs while attention costs grow quickly. Infini-attention processes local segments and carries older information through a compressed memory, reducing the need to recompute the full history.
- [Mixture of Experts explained simply](https://www.louisbouchard.ai/moe/) - Published Apr 09, 2024. Updated Jan 23, 2025. Tags: Large Language Models. Mixtral MoE Explained. This means that the model tries to predict the next token, or next word, of a sentence you send as the input prompt. Questions: What is a mixture-of-experts language model? Why use experts instead of activating the whole model? What does the router do? Takeaways: This means that the model tries to predict the next token, or next word, of a sentence you send as the input prompt. First, obviously, you’ll have your text and need to get your embeddings, which are the numbers the model understands.
- [AI in Education: The End of Schools As We Know Them?](https://www.louisbouchard.ai/ai-in-education/) - Published Mar 25, 2024. Updated Mar 25, 2024. Tags: Learning AI. AI's Impact on Traditional Education. This technology will change not only how and what we learn but also the skills we need for the future. Questions: AI in Education: The End of Schools As We Know Them? How could AI change what students need to learn? How can students use AI without outsourcing the learning? Takeaways: This technology will change not only how and what we learn but also the skills we need for the future. Once we understand how it works and how to write, we can use Antidote to double-check us.
- [How will AI impact Healthcare?](https://www.louisbouchard.ai/healthcare/) - Published Mar 12, 2024. Updated Mar 12, 2024. Tags: Large Language Models. A practical look at where AI already helps healthcare, from medical imaging and monitoring to drug research, plus the privacy, bias, and safety limits. Questions: How will AI impact Healthcare? Where is AI already used in healthcare? What is federated learning in healthcare? Takeaways: AI already supports medical imaging, digital stethoscopes, wearable monitoring, drug research, administrative work, and robotic-assisted procedures. Healthcare data is sensitive and fragmented. Federated learning can help hospitals train shared models without pooling the raw patient data in one place.
- [AI Revolution in Journalism: What It Means for News](https://www.louisbouchard.ai/ai-and-journalism/) - Published Feb 26, 2024. Updated Feb 26, 2024. Tags: Large Language Models, AI News and Analysis. Will journalists be replaced by AI? You don’t need much other than ChatGPT or another similar language model like Gemini. Questions: Which journalism tasks can current AI tools accelerate? Will AI remove the need for journalists? Why is real-time transcription useful in a newsroom? Takeaways: You don’t need much other than ChatGPT or another similar language model like Gemini. We think of an idea or story to share and, on the opposite, skip words with no problem.
- [Get Hired in Data Analytics Within 90 Days: Avery Smith's Success Formula](https://www.louisbouchard.ai/avery-smith-data-job/) - Published Feb 12, 2024. Updated Feb 12, 2024. Tags: Learning AI. Avery Smith shares a practical route into data analytics: build real projects, learn the tools employers use, and prove you can turn data into decisions. Questions: Can someone enter data analytics within 90 days? Which projects make a data analytics portfolio credible? What tools should an aspiring analyst prioritize? Takeaways: A portfolio built around real business problems says more than a long list of courses, especially for career changers without a technical degree. Start with Excel, SQL, and a visualization tool such as Power BI or Tableau, then add Python when the role or project actually needs it.
- [Tina Huang on AI, Education, Freelancing, and Boosting Personal Productivity](https://www.louisbouchard.ai/tina-huang/) - Published Feb 05, 2024. Updated Feb 05, 2024. Tags: Learning AI. Tina Huang shares practical advice on learning AI, choosing education, building projects, freelancing, and using AI tools without outsourcing judgment. Questions: What does the conversation with Tina Huang cover? Why does problem solving matter more as tools improve? Are certificates useful for an AI career? Takeaways: A degree can provide structure and still matters for some paths, but projects show whether you can apply the skills to a real problem. One certificate can signal basic preparation when you have no related background. Collecting ten certificates does not replace practical work.
- [AI and Art - Redefining Creativity with Mariam Brian](https://www.louisbouchard.ai/art-and-ai/) - Published Jan 30, 2024. Updated Jan 30, 2024. Tags: AI Engineering. A conversation with Holo Art CEO Mariam Brian about AI as a creative tool, artistic ownership, ethics, and how artists may work with models. Questions: Does using AI make someone less creative? What should artists check before using a generative model? Why does artistic ownership become difficult with AI? Takeaways: AI can act as a creative tool and collaborator, but the artist still supplies the intent, judgment, and cultural context. The difficult questions are ownership, consent, authenticity, and whether artists can say no when their work is used to train models.
- [ChatGPT: Understanding Its Power Through RLHF and RLAIF](https://www.louisbouchard.ai/rlhf-et-rlaif/) - Published Jan 23, 2024. Updated Jan 23, 2024. Tags: RLHF, RLAIF, Large Language Models. Understand how RLHF uses human preferences to align language models and how RLAIF automates part of that process with feedback from another AI model. Questions: What does RLHF mean? At what stage of training is RLHF used? How are human preferences collected? Takeaways: RLHF turns human rankings into a reward model that then guides language model training. This stage follows pre-training and supervised fine-tuning to move model responses closer to human preferences.
- [How is Data Transforming AI](https://www.louisbouchard.ai/jerome-pasquero/) - Published Jan 22, 2024. Updated Jan 22, 2024. Tags: AI Engineering. Jerome Pasquero on Data Annotation: The Building Blocks of AI's Future - The What's AI Podcast (episode 27). Questions: Why is data annotation important for AI? How are large annotated datasets built? When should humans annotate instead of AI? Takeaways: You know, a kid doesn't need that much data to learn the difference between a cat and a dog, right? We cover what's it used for, how do we build those big data sets, the difference between using AI or humans, how many humans do you need, and much more details.
- [AI Unveiled: Ethics, Innovation, and the Human-AI Symbiosis](https://www.louisbouchard.ai/jeremy-cohen-2/) - Published Jan 15, 2024. Updated Jan 15, 2024. Tags: AI Engineering. Jérémy Cohen discusses autonomous vehicles, fragile AI startups, hallucinations, bias, and why transparency and human judgment still matter. Questions: What themes does the Jérémy Cohen conversation explore? Will autonomous vehicles automatically eliminate traffic? Why are some AI startups fragile? Takeaways: Autonomous vehicles may improve safety, but adding more cars does not automatically solve congestion or weak transportation infrastructure. An AI startup that depends entirely on one model provider gives up control and is easy for the platform, or another thin wrapper, to replace.
- [RAG Explained](https://www.louisbouchard.ai/rag-explained/) - Published Jan 09, 2024. Updated Jan 09, 2024. Tags: RAG and Retrieval, Generative AI. Retreival augmented generation (RAG) explained simply. If you need safe information from an AI chatbot like a medical assistant, a tutor, a lawyer. Questions: What is retrieval-augmented generation? What data can a RAG system use? How does RAG reduce hallucinations? Takeaways: If you need safe information from an AI chatbot like a medical assistant, a tutor, a lawyer, or an accountant, you will be using RAG for sure. I hope you’ve enjoyed this video and that it helped you understand the goals and principles of RAG better.
- [2023: The best AI papers - A Review 🚀](https://www.louisbouchard.ai/2023-the-best-ai-papers-a-review/) - Published Dec 24, 2023. Updated Dec 24, 2023. Tags: Large Language Models, AI News and Analysis. Lastly, we witnessed controversy surrounding Google’s demonstration video of the Gemini model and more… In 2022, we witnessed the rise of the mighty ChatGPT. Questions: What does this 2023 AI recap focus on? What limitations still affected image models in 2023? Why did Google's Gemini demonstration become controversial?
- [What's AI Episode 25: Jerry Liu. From RAG Strategies to Gemini's Impact in Tech](https://www.louisbouchard.ai/jerry-liu/) - Published Dec 19, 2023. Updated Dec 19, 2023. Tags: Large Language Models, RAG and Retrieval, AI Engineering. It's just a programmatic way of prompt engineering. if we think RAG will get better than that, we could see that as a way of prompt engineering. Questions: What does LlamaIndex help developers build? Why describe RAG as programmatic prompt engineering? When is a vector database useful?
- [How to have a successful AI startups](https://www.louisbouchard.ai/greg-coquillo/) - Published Dec 11, 2023. Updated Dec 11, 2023. Tags: Learning AI. Lessons on building AI startups, finding useful problems, and avoiding the usual hype traps from Greg Coquillo's founder perspective. Questions: What makes an AI startup problem worth solving? How much industry knowledge does an AI founder need? Why is customer selection as important as the model?
- [Train, fine-tune, prompt, RAG... What should you do with your LLM?](https://www.louisbouchard.ai/train-fine-tune-prompt-rag/) - Published Nov 30, 2023. Updated Nov 30, 2023. Tags: Large Language Models, RAG and Retrieval, AI Engineering. Tips on what to do with your language model or API. Do you ever question yourself if you should be training from scratch, fine-tuning. Questions: Train, fine-tune, prompt, RAG... What should you do with your LLM? When should I start with prompting? When should I add RAG?
- [The Future of Video Generation: Deep Dive into Stable Video Diffusion](https://www.louisbouchard.ai/stable-video-diffusion/) - Published Nov 28, 2023. Updated Nov 28, 2023. Tags: Computer Vision, Generative AI. Stable Video Diffusion Explained. I’m Louis from What’s AI, and let’s explain how this new model works! Questions: What is Stable Video Diffusion? Why generate video in a compressed latent space? How does the model return to visible frames?
- [Understanding Distil-Whisper: A Closer Look at AI's Role in Audio Transcription](https://www.louisbouchard.ai/distil-whisper/) - Published Nov 23, 2023. Updated Nov 23, 2023. Tags: Generative AI. It’s a much more complex problem than processing text as you also need to deal with background noises, different tones or pitch, and, obviously. Questions: What is Distil-Whisper? Why is audio transcription harder than text processing? How is the distilled model trained?
- [AI Expert Ken Jee Discusses Data Science & AI Startups](https://www.louisbouchard.ai/ken-jee/) - Published Nov 21, 2023. Updated Nov 21, 2023. Tags: Learning AI. Data Science, Startups & More with Ken Jee. You need to be passionate about the problem and try to find solutions, whatever it is. Questions: What does the conversation with Ken Jee cover? Why should data scientists focus on the problem before the tool? How can personal interests support a data portfolio?
- [Start with Large Language Models (LLMs) in 2023 - Become an expert for free!](https://www.louisbouchard.ai/start-with-large-language-models-llms-in-2023-become-an-expert-for-free/) - Published Nov 17, 2023. Updated Nov 17, 2023. Tags: Large Language Models, Learning AI, AI News and Analysis. Learn LLMOps now. A complete guide. Free. LLMs are revolutionizing how we interact with technology, from chatbots to advanced data analysis. Questions: What should a beginner learn first about LLMs? Why learn LLMOps? Which project is a useful introduction to RAG?
- [Revolutionizing AI Applications: Dmitry Shapiro's Vision for MindStudio](https://www.louisbouchard.ai/revolutionizing-ai-applications-dmitry-shapiros-vision-for-mindstudio/) - Published Nov 15, 2023. Updated Nov 16, 2023. Tags: Computer Vision, Generative AI, AI News and Analysis. This is an interview with Dmitry Shapiro, previously working at Google and CTO of MySpace Music. Now. Questions: What is MindStudio? What does the interview with Dmitry Shapiro cover? Why should model selection follow the use case?
- [Navigating the AI Landscape: Strategies from Paige Bailey, AI Product Leader](https://www.louisbouchard.ai/paige-bailey/) - Published Nov 09, 2023. Updated Nov 09, 2023. Tags: Large Language Models, AI News and Analysis. From Microsoft GitHub to Google DeepMind: Paige Bailey. And also you should probably be doing this in the context of an IDE like Android Studio. Questions: What does the interview with Paige Bailey cover? Why should coding assistance live inside an IDE? How do feedback loops improve an AI product?
- [I dropped out of my PhD](https://www.louisbouchard.ai/i-quit-my-phd/) - Published Nov 05, 2023. Updated Dec 01, 2023. Tags: Large Language Models, Learning AI. Dropping out of a PhD for the startup world. Before the Decision… Just to give some context, my journey into the Ph.D. was quite obvious to me. Questions: Why did I decide to leave my AI PhD? Was leaving the PhD an easy decision? What is the sunk-cost fallacy in this context?
- [The Future of AI: 5 Incredible Ways GPT-4 is Shaping Our World](https://www.louisbouchard.ai/5-apps-gpt-4/) - Published Oct 26, 2023. Updated Oct 26, 2023. Tags: Large Language Models. Discover 5 useful Applications of GPT-4 and Llama-2 - No Coding Required! This might not be super useful in the real world. Questions: Do these GPT-4 applications require coding experience? How can an LLM work with private documents or other data sources? Why would a brand give a language model a specific persona?
- [Meet Our new AI Tutor!](https://www.louisbouchard.ai/rag-aitutor/) - Published Oct 25, 2023. Updated Oct 25, 2023. Tags: Large Language Models, RAG and Retrieval, AI Engineering, Generative AI. How we Built an Open-Source RAG-based ChatGPT Web App. That’s where the marvel of Retrieval Augmented Generation, or RAG, comes into play. Questions: What is the open-source AI tutor? Why use RAG for a tutor? What happens when the knowledge base lacks an answer?
- [DALLE 3 Explained: Improving Image Generation with Better Captions](https://www.louisbouchard.ai/dalle-3/) - Published Oct 21, 2023. Updated Nov 15, 2023. Tags: Computer Vision, Generative AI, AI News and Analysis. OpenAI’s BEST text-to-image model to date! Or, we could use another model to understand images and generate better captions first! Questions: Why did better captions improve DALL-E 3? What was missing from many original image captions? How were stronger training captions produced?
- [Our AI Tutor is live!!!](https://www.louisbouchard.ai/aitutor/) - Published Oct 17, 2023. Updated Oct 19, 2023. Tags: Large Language Models, RAG and Retrieval, Learning AI. Last week, we released our LLM course. We are excited to announce another related project: our own AI tutor! Questions: What is the AI tutor designed to answer? How does RAG make the tutor more useful? Who is the accompanying LLM course for?
- [Responsible AI: Insights from an Ethics Expert](https://www.louisbouchard.ai/responsible-ai/) - Published Oct 06, 2023. Updated Oct 06, 2023. Tags: AI Engineering. The What's AI Podcast Episode 20 with Auxane Boch: Ethics Expert. And you, you check all those boxes to be sure that there is a problem and that your fix. Questions: What is responsible AI in practice? Why start with the problem rather than the API? How can a team identify a high-risk AI use case?
- [Mastering Prompts: How to Effectively Communicate with AI Language Models](https://www.louisbouchard.ai/prompting-llms/) - Published Sep 29, 2023. Updated Sep 29, 2023. Tags: Large Language Models. Exploring the unique behaviors of different Large Language Models (LLMs) and mastering advanced prompting techniques! Questions: Why do different LLMs respond differently to the same prompt? Which generation settings should users test? How can a prompt request simpler language?
- [Generate music with AI: Stable Audio Explained](https://www.louisbouchard.ai/stableaudio/) - Published Sep 27, 2023. Updated Sep 27, 2023. Tags: Computer Vision, Generative AI. New AI That Generates useful Music! This is used to allow the model to understand how long the song generated should be and understand various lengths as well. Questions: What does Stable Audio generate? How does the model control audio length? Why use a latent audio representation?
- [Adapting LLMs to do specific tasks!](https://www.louisbouchard.ai/fine-tuning/) - Published Sep 19, 2023. Updated Sep 19, 2023. Tags: Large Language Models. Boost AI Performance with Fine-Tuning. This means re-training parts of the model on the exact task we want it to achieve. Questions: What is LLM fine-tuning? When is fine-tuning more useful than prompting? Can fine-tuning teach a model to admit uncertainty?
- [MVDream: Creating Lifelike 3D Models from Words](https://www.louisbouchard.ai/mvdream/) - Published Sep 10, 2023. Updated Sep 10, 2023. Tags: Computer Vision. MVDream: A new text-to-3d approach (explained)! We do all that to better understand where in the image the NeRF model should focus its attention to produce. Questions: What does MVDream generate? Why generate multiple views together? How is a multi-view diffusion model different from text-to-image?
- [AI Deep Learning Explained](https://www.louisbouchard.ai/ai-deep-learning-explained/) - Published Sep 06, 2023. Updated Sep 06, 2023. Tags: Learning AI. Deep learning with a simple analogy. They are powerful because they will have excellent results if you have the data to teach them what you want. Questions: What is deep learning in simple terms? Why do deep learning systems need so much data? What is supervised learning?
- [The First General-Purpose Visual and Language AI: LLaVA](https://www.louisbouchard.ai/llava/) - Published Sep 02, 2023. Updated Jan 09, 2024. Tags: Large Language Models, Computer Vision. LLaVA: Bridging the Gap Between Visual and Language AI with GPT-4. We train the LLaMA model and the image-to-text conversion. Questions: What is LLaVA? How does LLaVA connect images to language? Why must the model inspect the image itself?
- [Fixing AI Hallucinations: The Importance of Explainability in AI (XAI)](https://www.louisbouchard.ai/xai-2/) - Published Aug 30, 2023. Updated Aug 30, 2023. Tags: Computer Vision. An introduction to explainable AI. You are a research scientist aiming to build a model able to classify images. Questions: What is explainable AI? How can an attention or saliency map reveal a failure? Why is confident accuracy not enough?
- [Mitigating AI Hallucinations: Exploring MetaGPT's Collaborative Framework](https://www.louisbouchard.ai/metagpt/) - Published Aug 27, 2023. Updated Aug 27, 2023. Tags: AI Agents, Large Language Models, AI News and Analysis. What is MetaGPT? LLM Agents Collaborating to Solve Complex Tasks. The goal of MetaGPT is simple: replace each with a fictive human being: a GPT model. Questions: What is MetaGPT? How is it different from a single autonomous agent? Why might role separation reduce some errors?
- [Large Language Models Enter the 3D World!](https://www.louisbouchard.ai/3d-llm/) - Published Aug 11, 2023. Updated Aug 11, 2023. Tags: Large Language Models, Computer Vision. An overview of the first 3D-LLM. Usually, just to match images to text, you need lots of examples of text-image pairs so that the model can understand both. Questions: What is a 3D large language model? Why can’t a standard image-language model directly understand a 3D scene? What data does a 3D-LLM need for training?
- [StyleGANEX: Enhancing Image Manipulation with Dilated Convolutions](https://www.louisbouchard.ai/styleganex/) - Published Aug 03, 2023. Updated Aug 03, 2023. Tags: Computer Vision, Generative AI. StyleGANEX: Enhancing Image Manipulation with Dilated Convolutions explains how visual AI systems work, where they break. Questions: What problem does StyleGANEX address? Why does ordinary StyleGAN editing require aligned faces? What do dilated convolutions change?
- [Leaving Deepmind to start his own Startup! Aleksa Gordić](https://www.louisbouchard.ai/aleksa/) - Published Jul 12, 2023. Updated Jul 12, 2023. Tags: Learning AI, AI News and Analysis. Aleksa Gordić - The What's AI Podcast Episode 18. Having said that, like videos are obviously so much more data intensive, like processing wise, you. Questions: What is this podcast episode with Aleksa Gordić about? How did Aleksa build practical experience outside formal coursework? Why can writing a technical blog help an AI career?
- [Google Maps Travel Time Prediction Algorithm & AI Research at Google Deepmind](https://www.louisbouchard.ai/petar/) - Published Jul 04, 2023. Updated Jul 04, 2023. Tags: AI News and Analysis. An interview with Petar Veličković - The What's AI Podcast Episode 17. Could you give some, not practical but applied types of data that this can work with. Questions: What does the interview with Petar Veličković cover? Why are graphs useful for travel-time prediction? What should a reasoning system handle as inputs grow?
- [Building LLM Apps and the Challenges that come with it: An Interview with Jay Alammar](https://www.louisbouchard.ai/jay-alammar/) - Published Jun 27, 2023. Updated Jun 27, 2023. Tags: Large Language Models. The What's AI podcast episode 16 with Jay Alammar. Break it down into sentences, embed each sentence, and then you can [00:36:00] create. Questions: What does the Jay Alammar interview cover? Why split documents before embedding them? What can embeddings add to an application?
- [Revolutionizing Online Shopping: AI's Virtual Try-On Experience](https://www.louisbouchard.ai/tryondiffusion/) - Published Jun 24, 2023. Updated Jun 24, 2023. Tags: Computer Vision, Generative AI. AI Does magic with UNets, Diffusion, and clothes! It’s also perfectly fine if you are not into podcasts and just want to understand how the TryOnDiffusion. Questions: What does TryOnDiffusion generate? What must the model understand from both inputs? Why use diffusion for virtual try-on?
- [What is Active Learning in AI?](https://www.louisbouchard.ai/active-learning/) - Published Jun 17, 2023. Updated Jun 17, 2023. Tags: Learning AI. Active learning explained in 5 minutes. It means we need lots of very good, balanced, and varied data. Questions: What is Active Learning in AI? What is active learning in AI? How does a model choose which examples need labels?
- [Luis Serrano: Exploring LLMs and the Journey of an AI Scientist](https://www.louisbouchard.ai/luis-serrano/) - Published Jun 15, 2023. Updated Jun 15, 2023. Tags: Large Language Models. What's AI Podcast Episode 15 with Luis Serrano from Cohere. And right now, since they are not perfect. Questions: What does the Luis Serrano interview explore? Should beginners learn all the mathematics before building a model? Why can an LLM be explained through small components?
- [The Best Image-to-3D AI to date: Neuralangelo](https://www.louisbouchard.ai/neuralangelo/) - Published Jun 11, 2023. Updated Jun 11, 2023. Tags: Computer Vision. Great, now that you know about Instant NGP, let’s look at the two key differences in improving the results with Neuralangelo working on this specific hash. Questions: What does Neuralangelo reconstruct? How is it related to Instant NGP? Why blend information across neighboring grid cells?
- [From GPT3 to AGI: Insights from Felix Tao, CEO of Mindverse AI](https://www.louisbouchard.ai/from-gpt3-to-agi-insights-from-felix-tao-ceo-of-mindverse-ai/) - Published Jun 06, 2023. Updated Jun 06, 2023. Tags: Large Language Models. Navigating the Changing Landscape of AI: Felix Tao's Journey from Researcher to CEO. Questions: What does the conversation with Felix Tao cover? Why might a domain model need reasoning more than broad knowledge? What is an AI self-awareness module in this discussion?
- [Image Manipulation with Your Mouse! Drag Your Gan Explained](https://www.louisbouchard.ai/draggan/) - Published May 27, 2023. Updated May 27, 2023. Tags: RAG and Retrieval, AI Engineering, Computer Vision, Generative AI. Move anything from point A to point B! This kind of architecture is a way to train a generative model where you will train a model, called your generator. Questions: What does DragGAN let a user do? Why not move the pixels directly? What is a handle point?
- [What is Explainability AI?](https://www.louisbouchard.ai/xai/) - Published May 25, 2023. Updated May 25, 2023. Tags: AI Engineering. Interview with Yotam Azriel, CTO at TensorLeap - What's AI episode 13. 36:30 Can we do anything to understand the results of the model better when we work. Questions: What is Explainability AI? What does the interview with Yotam Azriel cover? Why fine-tune an ImageNet model instead of training from scratch?
- [What is Prompt Injection? "Prompt Hacking" Explained](https://www.louisbouchard.ai/prompt-injection/) - Published May 23, 2023. Updated May 23, 2023. Tags: Large Language Models. By now, you all know what prompting is. It’s how we talk with ChatGPT and other AIs. Questions: What is prompt injection? How is prompt injection different from an ordinary prompt? Why is injection dangerous for tool-using models?
- [Copilot, ChatGPT and Copyrights](https://www.louisbouchard.ai/copilot-chatgpt-and-copyrights/) - Published May 16, 2023. Updated May 16, 2023. Tags: Large Language Models, Generative AI. Something we need to be careful when it comes to plagiarism and arises new questions around the attribution and moral rights of such generated pieces. Questions: Why do coding assistants raise copyright questions? Can AI-generated code still count as plagiarism? What technical problems can generated code contain?
- [Your Dictionary for AI in 2023](https://www.louisbouchard.ai/ai-dictionary/) - Published May 11, 2023. Updated May 11, 2023. Tags: Large Language Models, Computer Vision, Generative AI, AI News and Analysis. Logan Kilpatrick's Definitions for all the words to know for OpenAI, GPT and AI in 2023. Questions: Why is an AI dictionary useful for beginners? What does alignment mean for a language model? Why is a newly trained base model not immediately ready for users?
- [An Interview with OpenAI's Developer Advocate Logan Kilpatrick](https://www.louisbouchard.ai/logan-kilpatrick-openai/) - Published May 09, 2023. Updated May 09, 2023. Tags: Large Language Models, Computer Vision, Generative AI, AI News and Analysis. Building with LLMs, ChatGPT, and Working at OpenAI With Logan Kilpatrick (Dev Rel @OpenAI) - What's AI episode 11. Questions: What does the interview with Logan Kilpatrick cover? What does a developer advocate do? Why were ChatGPT plugins important to developers?
- [Perfusion: Stable Diffusion but more Controllable!](https://www.louisbouchard.ai/perfusion/) - Published May 05, 2023. Updated May 05, 2023. Tags: Computer Vision, Generative AI, AI News and Analysis. The new Perfusion model by NVIDIA Explained. Oh, and how can we do that without having an immense model that only OpenAI can use? Questions: What does NVIDIA's Perfusion personalize? What two objectives must personalization balance? How is Perfusion more compact than full fine-tuning?
- [How AI Enhances Writing & Affects Journalism](https://www.louisbouchard.ai/ai-writing/) - Published Apr 26, 2023. Updated Apr 26, 2023. Tags: AI Engineering. Join us for an engaging podcast episode with Limarc Ambalina, the Vice President of Editing at HackerNoon. Questions: What is the difference between AI-generated and AI-edited writing? Why does transparency matter when AI is used in writing? How can AI improve a human-written draft?
- [Break into AI in 2023: Best Resume & Interview Tips with Brian Burns (AI Pub) - What's AI Podcast Episode 9](https://www.louisbouchard.ai/brian-burns/) - Published Apr 19, 2023. Updated Apr 19, 2023. Tags: AI Engineering. Ph.D., Twitter, Breaking into the AI Field, and more with Brian Burns (AI Pub). Not really, you will also most probably need to make money. Questions: Can someone enter AI without a machine learning PhD? When is a PhD useful for an AI career? What should an AI resume emphasize?
- [10x Your Productivity with those AI Tools](https://www.louisbouchard.ai/ai-tools-productivity/) - Published Apr 16, 2023. Updated Apr 16, 2023. Tags: AI Engineering. These AI Tools will 10x Your Productivity! I’ve used it since we had it for free in my first data science job in 2020 and have used it ever since. Questions: Which kinds of work can these AI tools speed up? Why is summarization useful in a productivity workflow? What does Otter.ai add to meetings?
- [What is a Self-Driving Car Engineer?](https://www.louisbouchard.ai/selfdrivingcars/) - Published Apr 11, 2023. Updated Apr 12, 2023. Tags: AI Engineering. Insights to become a self-driving car engineer in 2023. Have you ever wondered what it takes to work in the field of self-driving cars? Questions: What is a Self-Driving Car Engineer? What does a self-driving car engineer work on? Which skills are useful for autonomous-driving ML?
- [Meta's new Segment Anything Model Explained](https://www.louisbouchard.ai/meta-sam/) - Published Apr 06, 2023. Updated Apr 06, 2023. Tags: Large Language Models, Computer Vision, AI News and Analysis. SAM, promptable segmentation and the largest segmentation dataset to date! This means you can help it segment unknown objects through prompts without. Questions: What does the Segment Anything Model do? What is promptable segmentation? Why is segmentation useful?
- [How to talk with AIs with Learn Prompting's creator](https://www.louisbouchard.ai/how-to-talk-with-ais-with-learn-promptings-creator/) - Published Mar 31, 2023. Updated Apr 03, 2023. Tags: Large Language Models, Learning AI. A conversation with Learn Prompting's creator on prompting, AI literacy, and how people can communicate better with models. Questions: What is a prompt? What is few-shot prompting? What is zero-shot chain-of-thought prompting?
- [Google’s New AI Robot Can See and Understands Language! (PaLM-E)](https://www.louisbouchard.ai/palm-e/) - Published Mar 23, 2023. Updated Mar 23, 2023. Tags: AI News and Analysis. Since this PaLM model is already super powerful to understand and interact with the world in a textual format, the only thing we need to do is to make it see. Questions: What is PaLM-E? How does a language model learn to see? How does PaLM-E connect understanding to action?
- [What is a Solution Architect at NVIDIA?](https://www.louisbouchard.ai/solution-architect/) - Published Mar 19, 2023. Updated Mar 19, 2023. Tags: AI News and Analysis. Here’s an interview with Adam Grzywaczewski, a senior deep learning architect at NVIDIA. Questions: What is a Solution Architect at NVIDIA? What does a solution architect at NVIDIA do? How is solution architecture different from pure research?
- [How good is GPT-4?](https://www.louisbouchard.ai/gpt-4/) - Published Mar 15, 2023. Updated Mar 15, 2023. Tags: Large Language Models. GPT-4 may be the most hyped language model we’ve had, with tons of rumors and news even prior to its announcement, and it wasn’t for nothing. Questions: How good is GPT-4? What improved with GPT-4? What is the purpose of a system message?
- [How to Build a strong Data Science Resume with Kaggle](https://www.louisbouchard.ai/build-a-strong-data-science-resume-with-kaggle/) - Published Mar 09, 2023. Updated Mar 09, 2023. Tags: AI News and Analysis. Use Kaggle projects to make a stronger data science resume, with practical proof of skills instead of generic claims. Questions: How can Kaggle strengthen a data science resume? Do competition rankings matter more than project explanations? What should a Kaggle portfolio entry include?
- [Interview with NVIDIA Data Scientist Meriem Bendris & RTX 4080 Giveaway!](https://www.louisbouchard.ai/nvidia-meriem-bendris/) - Published Mar 07, 2023. Updated Mar 07, 2023. Tags: AI News and Analysis. In this interview, we'll hear from Meriem Bendris, a senior solution architect at NVIDIA, who will answer questions such as:. Questions: Who is Meriem Bendris? What does the interview discuss? Who is the episode useful for?
- [Gen-1, the future of storytelling?](https://www.louisbouchard.ai/gen-1/) - Published Mar 06, 2023. Updated Mar 07, 2023. Tags: Computer Vision, Generative AI, Learning AI. The company behind one of the most popular image generation models you certainly know, Stable Diffusion, is taking it a step further. Questions: Gen-1, the future of storytelling? What does Runway Gen-1 do? Why not apply image diffusion independently to every frame?
- [Announcing the What's AI Podcast!](https://www.louisbouchard.ai/podcast-announcement/) - Published Mar 01, 2023. Updated Mar 01, 2023. Tags: AI Engineering. Introducing my podcast on AI! We cover what it is like to be a data scientist in such a big and important company and how to get there. Questions: What is the What's AI Podcast? Who appears in the first announced episode? Why add a podcast to the existing AI content?
- [What's it like to work in Generative AI?](https://www.louisbouchard.ai/generative-ai-startup/) - Published Feb 20, 2023. Updated Feb 20, 2023. Tags: Generative AI, Learning AI. An interview with the VP of R&D at a Generative AI company (D-ID). More specifically, we cover three main parts; education, the job interview process. Questions: What's it like to work in Generative AI? Do you need a master's degree to work in generative AI? What does generative AI research look like inside a startup?
- [Generating music with AI!](https://www.louisbouchard.ai/musiclm/) - Published Jan 31, 2023. Updated Jan 31, 2023. Tags: Generative AI, AI News and Analysis. MusicLM explained. This is how Google Research was able to create a model generating music with such great results from simple text descriptions! Questions: What does MusicLM generate? How is text-to-music different from text-to-speech? Can the prompt control the generated sound?
- [Image Editing from Text Instructions: InstructPix2Pix](https://www.louisbouchard.ai/instructpix2pix/) - Published Jan 26, 2023. Updated Jan 26, 2023. Tags: Computer Vision, Learning AI. This new model called InstructPix2Pix does exactly that; it edits an image following a text-based instruction given by the user. Questions: What does InstructPix2Pix do? What two conditions guide the output? How was the training dataset created?
- [VALL-E: An AI Generating Voice from Text!](https://www.louisbouchard.ai/vall-e/) - Published Jan 16, 2023. Updated Jan 16, 2023. Tags: AI Engineering. The VALL-E Model explained. For that, we need to translate it into something our model can understand, which is often matrices and numbers, not words or sounds. Questions: What is Microsoft's VALL-E? How does the model represent speech? What does the voice prompt provide?
- [An interview with the Director of Perception at Zoox: Ruijie (RJ) He: What is an ML Engineer and more...](https://www.louisbouchard.ai/rj-he-zoox/) - Published Jan 10, 2023. Updated Jan 10, 2023. Tags: AI Engineering. An interview with the Director of Perception at Zoox, Ruijie (RJ) He, with the goal of demystifying what is a good profile to get an ML engineer job and. Questions: What does a perception team do at Zoox? What does the interview with RJ He cover? What makes a strong machine-learning engineering profile?
- [The 2022 AI Recap - A Review 🚀](https://www.louisbouchard.ai/2022-ai-recap/) - Published Dec 23, 2022. Updated Dec 31, 2022. Tags: AI News and Analysis. A month-by-month review of the AI research and releases that shaped 2022, from image and video generation to language models and 3D systems. Questions: Which AI developments defined 2022? Why did generative AI move so quickly in 2022? What is the best way to use this 2022 recap? Takeaways: 2022 pushed generative AI into public view through rapid progress in image generation, language models, video, audio, and 3D. The strongest demos came from different model families, so compare how each system was trained, controlled, and evaluated.
- [Automatic Re-Aging with AI! Disney’s FRAN Model Explained](https://www.louisbouchard.ai/disney-re-age/) - Published Dec 22, 2022. Updated Dec 28, 2022. Tags: AI Engineering. Disney's New Model Explained. This is a big deal for the film industry, allowing you to instantly re-age someone for a whole movie with very little cost. Questions: What does Disney's FRAN model do? Why is consistent re-aging across frames difficult? How does FRAN limit the area it edits?
- [Prompting Explained: How to talk to ChatGPT](https://www.louisbouchard.ai/prompting-explained/) - Published Dec 13, 2022. Updated Dec 13, 2022. Tags: Large Language Models, AI Engineering. Learn what prompting and prompt engineering are, how text and image prompts work, and how to get better results from generative AI. Questions: What is prompting? What does a prompt engineer do? How can a beginner improve a prompt?
- [What is ChatGPT?](https://www.louisbouchard.ai/chatgpt/) - Published Dec 06, 2022. Updated Dec 06, 2022. Tags: Large Language Models, Learning AI, AI News and Analysis. OpenAI's most recent conversational AI explained. The second step is to add our reinforcement learning magic. Questions: What is ChatGPT? How did ChatGPT differ from the original GPT-3? Why was reinforcement learning added?
- [From Audio to Talking Heads in Real-Time with AI! RAD-NeRF explained](https://www.louisbouchard.ai/rad-nerf/) - Published Dec 03, 2022. Updated Dec 03, 2022. Tags: Computer Vision, Generative AI. We’ve heard of deepfakes, we’ve heard of NeRFs, and we’ve seen these kinds of applications allowing you to recreate someone’s face and pretty much make him. Questions: What does RAD-NeRF generate? What training input does it require? How can audio animate a face?
- [How AI Understands Words](https://www.louisbouchard.ai/text-embedding/) - Published Nov 28, 2022. Updated Nov 29, 2022. Tags: Large Language Models, RAG and Retrieval. Text Embedding Explained. You’ve certainly heard of the most known and powerful language model: GPT-3. Questions: What is a text embedding? Why are embeddings useful? How are embeddings used in RAG? Takeaways: Text embeddings turn words, sentences, or documents into vectors so software can compare meaning instead of only matching keywords. Embeddings are the base layer behind semantic search, recommendations, clustering, duplicate detection, and many RAG systems.
- [Galactica: What is it and What Happened?](https://www.louisbouchard.ai/galactica/) - Published Nov 22, 2022. Updated Nov 22, 2022. Tags: AI News and Analysis. Galactica, Meta AI's most recent model: The AI Scientist. Tokenization basically means the way the model will see the data instead of words, math. Questions: Galactica: What is it and What Happened? What was Meta's Galactica model? What tasks was Galactica designed to perform?
- [InfiniteNature-Zero: Fly Into Your Pictures With AI!](https://www.louisbouchard.ai/infinitenature-zero/) - Published Nov 17, 2022. Updated Nov 17, 2022. Tags: Computer Vision, Generative AI. Have you ever imagined being able to take a picture and just magically explain it as if it would be a door to another world? Questions: How is InfiniteNature-Zero different from the earlier method? What does the model generate from one picture? Why reconstruct the initial image during training?
- [eDiffi explained: New SOTA Image Synthesis model!](https://www.louisbouchard.ai/ediffi/) - Published Nov 03, 2022. Updated Nov 04, 2022. Tags: Computer Vision, Generative AI, AI News and Analysis. NVIDIA's new model has better results, more control and more fidelity than DALLE and Stable Diffusion! Questions: What is eDiff-I? How can users control where subjects appear? Why is spatial control useful?
- [Diffusion models: Everything you need to know](https://www.louisbouchard.ai/diffusion-models/) - Published Nov 03, 2022. Updated Nov 03, 2022. Tags: Computer Vision, Generative AI. Here's every vision application Diffusion models were a game changer in 2022: image, text, video, 3D, and more! Questions: What is a diffusion model? Why are diffusion models important for image generation? How does a diffusion model use a text prompt? Takeaways: Diffusion models generate images by learning how to reverse noise into structure, which makes the denoising process the key idea. Text prompts guide the generation, but they do not remove the need to inspect artifacts, bias, data limits, and licensing questions.
- [AI Image Editing from Text! Imagic Explained](https://www.louisbouchard.ai/imagic/) - Published Oct 21, 2022. Updated Oct 24, 2022. Tags: Computer Vision, Generative AI. If you think the recent image generation models like DALLE or Stable Diffusion are cool, you just won’t believe how incredible this one is. Questions: What does Imagic do? Which inputs does Imagic require? Why optimize a text embedding for the source image?
- [3D Models from Text! DreamFusion Explained](https://www.louisbouchard.ai/dreamfusion/) - Published Oct 15, 2022. Updated Oct 15, 2022. Tags: Computer Vision, Generative AI. How AI generates 3d models from only text! But then, how can you generate a 3D render from text if the NeRF model only works with images? Questions: What does DreamFusion generate? How can a NeRF learn without real training photos? Why render the object from many angles?
- [OpenAI's Most Recent Model: Whisper (explained)](https://www.louisbouchard.ai/whisper/) - Published Oct 06, 2022. Updated Oct 06, 2022. Tags: AI News and Analysis. A good transcription tool that would accurately understand what you say and write it down. Questions: What is Whisper? Why was Whisper important? What can Whisper be used for? Takeaways: Whisper showed how strong speech recognition can become when trained on broad, messy, multilingual audio. Transcription quality depends on accents, noise, microphones, languages, domain words, and whether timestamps matter.
- [Make-a-video: The AI Film Maker!](https://www.louisbouchard.ai/make-a-video/) - Published Sep 29, 2022. Updated Sep 29, 2022. Tags: RAG and Retrieval, Generative AI. Generates videos from text! You need a model that understands the world in a better way and leverages this level of understanding to generate a coherent. Questions: What does Meta's Make-A-Video do? Does training require a massive paired text-video dataset? How are the first video frames generated?
- [General Video Recognition with AI](https://www.louisbouchard.ai/general-video-recognition/) - Published Sep 08, 2022. Updated Sep 08, 2022. Tags: AI Engineering. What does such a model understand when it sees such a picture or, even more complex, a video? Questions: What is general video recognition? Why is video harder to understand than an image? What must a video model learn first?
- [Guiding Stable Diffusion with your Images](https://www.louisbouchard.ai/imageworthoneword/) - Published Sep 02, 2022. Updated Sep 02, 2022. Tags: Computer Vision, Generative AI, Learning AI. Personalizing Text-to-Image Generation using Textual Inversion. If you haven’t watched my video yet, you should pause this one. Questions: What is textual inversion? How does it personalize Stable Diffusion? Why use a learned token instead of a long description?
- [One of the Most Challenging Tasks for AI](https://www.louisbouchard.ai/psg/) - Published Sep 01, 2022. Updated Sep 01, 2022. Tags: Generative AI. A New Challenging Task for AI: panoptic scene graph generation. Well, you need two things: a dataset and a powerful model. Questions: What is panoptic scene graph generation? How is this harder than object detection? What does a scene graph contain?
- [How Stable Diffusion works? Latent Diffusion Models Explained](https://www.louisbouchard.ai/latent-diffusion-models/) - Published Aug 27, 2022. Updated Dec 03, 2022. Tags: AI Engineering, Computer Vision, Generative AI. A High-Resolution Image Synthesis Architecture: Latent Diffusion. This is why you need hundreds of GPUs to train such a model and why you wait a few minutes. Questions: What is a latent diffusion model? Why use latent space for diffusion? How is latent diffusion different from pixel diffusion? Takeaways: Latent diffusion makes image generation cheaper by doing much of the denoising work in a compressed representation instead of raw pixels. The latent space is powerful because it keeps important visual structure while making generation faster and more practical.
- [Build Animatable 3D Models with AI](https://www.louisbouchard.ai/banmo/) - Published Aug 13, 2022. Updated Aug 13, 2022. Tags: Computer Vision. BANMo tackles a task we call articulated 3D shape reconstruction, which means it works with videos and pictures to model deformable objects. Questions: What does BANMo reconstruct? What does articulated 3D reconstruction mean? Why are animals difficult to reconstruct?
- [Produce Amazing Artworks with Text and Sketches!](https://www.louisbouchard.ai/make-a-scene/) - Published Jul 19, 2022. Updated Jul 19, 2022. Tags: Computer Vision, Generative AI. "Make-A-Scene": a fantastic blend between text and sketch-conditioned image generation. Questions: What does Make-A-Scene generate? Why combine text with a sketch? How is this different from ordinary text-to-image generation?
- [How OpenAI Reduces risks for DALL·E 2](https://www.louisbouchard.ai/how-openai-reduces-risks-for-dall-e-2/) - Published Jul 16, 2022. Updated Jul 16, 2022. Tags: Computer Vision, Generative AI, AI News and Analysis. DALL·E 2 Pre-Training Mitigations. You can see how such a model can become dangerous: allowing anyone to generate anything. Questions: Why can a text-to-image model create safety risks? How did OpenAI filter DALL-E 2 training data? Why filter the data before model training?
- [CVPR 2022 Best Paper Honorable Mention: Dual-Shutter Optical Vibration Sensing](https://www.louisbouchard.ai/cvpr-2022-best-paper/) - Published Jul 13, 2022. Updated Jul 13, 2022. Tags: Generative AI, AI News and Analysis. They reconstruct sound using cameras and a laser beam on any vibrating surface, allowing them to isolate music instruments, focus on a specific speaker. Questions: How can a camera reconstruct sound? Why use both rolling- and global-shutter cameras? What does the laser contribute?
- [What is data-centric AI?](https://www.louisbouchard.ai/data-centric-ai/) - Published Jul 08, 2022. Updated Jul 08, 2022. Tags: AI Engineering. The beginning of data-centric AI with data programming. Unfortunately, the more data you have, the better it is. Questions: What is data-centric AI? Why is data-centric AI useful? What is the difference between model-centric and data-centric AI? Takeaways: Data-centric AI shifts attention from only changing models to improving the data, labels, coverage, and feedback loops around them. Better examples, cleaner labels, and clearer edge cases can improve a system more than swapping to a larger model.
- [No Language Left Behind](https://www.louisbouchard.ai/no-language-left-behind/) - Published Jul 06, 2022. Updated Jul 06, 2022. Tags: AI News and Analysis. Translating 200 languages with a single model - Meta AI. These two models are only used to build the 200 paired-languages datasets we need to train the final. Questions: What is No Language Left Behind? Why are low-resource languages difficult for translation models? How were missing language pairs expanded?
- [How does dalle-mini work?](https://www.louisbouchard.ai/dalle-mini/) - Published Jun 16, 2022. Updated Jun 16, 2022. Tags: Computer Vision, Generative AI. Dalle mini is a free, open-source AI that produces useful images from text inputs. Here's how it works. Questions: How does dalle-mini work? What is DALL-E mini? How does text become an image?
- [Google Brain's Answer to Dalle-e 2: Imagen](https://www.louisbouchard.ai/google-brain-imagen/) - Published May 24, 2022. Updated May 24, 2022. Tags: Computer Vision, Generative AI, AI News and Analysis. An AI that creates photorealistic images from input text better than Dall-e 2! But before using this diffusion model. Questions: What is Google's Imagen model? Why does Imagen need a strong text encoder? What does the diffusion process do?
- [How Uber uses AI to serve you better](https://www.louisbouchard.ai/uber-deepeta/) - Published May 21, 2022. Updated May 21, 2022. Tags: AI Engineering. How can Uber deliver food and always arrive on time or a few minutes before? How do they match riders to drivers so that you can \ always\ find a Uber? All. Questions: What is Uber DeepETA? Why is ETA prediction hard? Why does ETA accuracy matter? Takeaways: Uber's ETA problem is a good example of AI working inside a real product where small errors matter to millions of users. The model has to combine maps, traffic, routes, demand, historical behavior, and live signals instead of relying on one simple feature.
- [Deepmind's new model Gato is amazing!](https://www.louisbouchard.ai/deepmind-gato/) - Published May 13, 2022. Updated May 13, 2022. Tags: AI Agents, AI News and Analysis. Gato: A single Transformer to RuLe them all! The first generalist RL agent using transformers! Questions: What is DeepMind's Gato? Why is Gato described as an agent? How can one model handle different tasks?
- [This is a BIG step for GANs! BlobGAN Explained](https://www.louisbouchard.ai/blobgan/) - Published May 13, 2022. Updated May 13, 2022. Tags: Computer Vision, Generative AI. A GAN model that uses simple blobs to manipulate objects in images…. Once we have good results. Questions: What is BlobGAN? How do blobs make image editing easier? Can BlobGAN edit objects independently?
- [Meta's new model OPT is GPT-3's closest competitor! (and is open source)](https://www.louisbouchard.ai/opt-meta/) - Published May 06, 2022. Updated May 06, 2022. Tags: Large Language Models, AI News and Analysis. An open-source model that is as powerful as GPT-3! You can say that this new model is very similar to GPT-3 but open-source and better for the environment. Questions: What is Meta's OPT-175B? How is OPT similar to GPT-3? Why did an open release matter?
- [Your Personal Photoshop Expert with AI!](https://www.louisbouchard.ai/mystyle/) - Published Apr 28, 2022. Updated Apr 28, 2022. Tags: Computer Vision. This AI can reconstruct, enhance and edit your images! It will basically find you in the encoded representation of all faces and be re-trained to push the. Questions: What does MyStyle personalize? Why does personalization improve face restoration? Which edits can the personalized model perform?
- [An AI that Automatically Summarizes your Documents](https://www.louisbouchard.ai/google-docs-summary/) - Published Apr 21, 2022. Updated Apr 21, 2022. Tags: Learning AI, AI News and Analysis. A new model for automatically generating summaries using machine learning, released in Google Docs that you can already use! Questions: What is AI document summarization? When is document summarization useful? What should a good AI summary include? Takeaways: AI document summarization is useful when it saves reading time without hiding the evidence the summary came from. A good summary should preserve the main decisions, constraints, and unanswered questions, not only compress the text.
- [OpenAI's new model DALL·E 2 is amazing!](https://www.louisbouchard.ai/openais-new-model-dall-e-2-is-amazing/) - Published Apr 07, 2022. Updated Apr 07, 2022. Tags: Computer Vision, Generative AI, AI News and Analysis. a useful new model by OpenAI capable of generating and inpainting images from a text input with incredible results! Questions: What can DALL-E 2 do? How does DALL-E 2 connect text and images? What does the image decoder do?
- [NVIDIA Turns Photos into 3D Scenes in Milliseconds | Instant NeRF Explained](https://www.louisbouchard.ai/nvidia-photos-into-3d-scenes/) - Published Mar 31, 2022. Updated Dec 03, 2022. Tags: Computer Vision, AI News and Analysis. Then, after such a training process, you only feed the images with the camera angles at inference time. Questions: What is Instant NeRF? What inputs does a NeRF need? Why are hash-grid encodings important?
- [Combine Lidar and Cameras for 3D object detection - Waymo](https://www.louisbouchard.ai/waymo-lidar/) - Published Mar 25, 2022. Updated Mar 26, 2022. Tags: Computer Vision, AI News and Analysis. Combine Lidar and regular Cameras for 3D object detection - Waymo & Google Research. Questions: Why combine LiDAR with cameras? How is LiDAR represented for the network? Why is the LiDAR pseudo-image sparse?
- [Impressive photo restoration by AI !](https://www.louisbouchard.ai/gfp-gan/) - Published Mar 11, 2022. Updated Jul 03, 2023. Tags: Computer Vision, Generative AI. This new and completely free AI model can fix most of your old pictures in a split second! Questions: What does GFP-GAN restore? Why use a pretrained StyleGAN2 model? What is a generative facial prior?
- [Top 5 AI Articles of February - Hackernoon](https://www.louisbouchard.ai/hn-top-5-articles-february/) - Published Mar 05, 2022. Updated Mar 05, 2022. Tags: AI Engineering. HackerNoon is the best place for software developers, blockchain experts, data scientists, and tech people like us to read, write, and publish. Questions: What topics appear in February's top five AI articles? Why learn reinforcement learning from a series? What does image color quantization demonstrate?
- [Speech Inpainting with AI !](https://www.louisbouchard.ai/speech-inpainting-with-ai/) - Published Feb 26, 2022. Updated Feb 28, 2022. Tags: Generative AI. So you just have to train this model once on a general dataset and then use it with your own audio tracks as it should, ideally. Questions: What is speech inpainting? How is speech inpainting similar to image inpainting? Can one trained model process new audio tracks?
- [Create Realistic 3D Renderings with AI !](https://www.louisbouchard.ai/neroic/) - Published Feb 12, 2022. Updated Feb 13, 2022. Tags: Computer Vision, Generative AI. From a few images to a 3D model using AI! And what if the lighting conditions vary in the pictures taken and the generated model looks different depending on. Questions: What does NeROIC reconstruct? Why are online image collections difficult to use? Why must a reconstructed object model lighting?
- [AI Facial Editing of Real Videos ! Stitch it in Time Explained](https://www.louisbouchard.ai/stitch-it-in-time/) - Published Jan 29, 2022. Updated Jan 29, 2022. Tags: AI Engineering. AI Facial Editing of Real Videos! Stitch it in Time Explained explains how visual AI systems work, where they break. Questions: What problem does Stitch It in Time solve? Why do independent frame edits look bad? Why crop and align the face?
- [Remove Undesired Objects From Your Images With Artificial Intelligence!](https://www.louisbouchard.ai/lama/) - Published Jan 15, 2022. Updated Jan 15, 2022. Tags: Computer Vision, Learning AI. Learn how this algorithm can understand images and automatically remove the undesired object or person and save your future Instagram post! Questions: What does LaMa do? Why is object removal difficult for a model? How is inpainting different from cropping?
- [Top 10 Computer Vision Papers of 2021](https://www.louisbouchard.ai/top-10-cv-papers-2021/) - Published Jan 01, 2022. Updated Jan 01, 2022. Tags: Computer Vision, AI News and Analysis. The top 10 computer vision papers in 2021 with video demos, articles, code, and paper reference. Questions: What does the 2021 computer-vision list include? Which major research themes appear? Why was DALL-E important to computer vision?
- [2021: A Year Full of Amazing AI papers - A Review](https://www.louisbouchard.ai/2021-ai-papers-review/) - Published Dec 25, 2021. Updated Dec 25, 2021. Tags: AI News and Analysis. A month-by-month review of the AI papers that shaped 2021, with short explanations, videos, articles, code, and full references. Questions: What does the 2021 AI papers review cover? Why was DALL-E an important text-to-image paper? What problem did VOGUE explore?
- [The AI Monthly Top 3 - December 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-december-2021/) - Published Dec 23, 2021. Updated Dec 23, 2021. Tags: AI News and Analysis. The most interesting December's AI breakthroughs with video demos, short articles, code, and paper reference. Questions: Which papers appear in the December 2021 roundup? What does EditGAN make controllable? What scale does CityNeRF handle?
- [New SOTA Image Captioning: ClipCap](https://www.louisbouchard.ai/clipcap/) - Published Dec 22, 2021. Updated Dec 22, 2021. Tags: Large Language Models, Computer Vision, Generative AI. Easily generate text descriptions for images using CLIP and GPT models! We could use such a model to describe movies automatically or create better. Questions: What does ClipCap do? What role does CLIP play? Why is a language model still required?
- [CityNeRF: 3D Modelling at City Scale!](https://www.louisbouchard.ai/citynerf/) - Published Dec 18, 2021. Updated Dec 18, 2021. Tags: Computer Vision, Generative AI. Generate city-scale 3D scenes with high-quality details at any scale! I won’t be covering how NeRF works since I have already done this in an article if you. Questions: What problem does CityNeRF solve? How large is the supported viewing range? What controls the level of detail?
- [Why should you sleep for 8 hours?](https://www.louisbouchard.ai/sleep-the-secret-for-productivity-and-a-better-life/) - Published Dec 11, 2021. Updated Dec 13, 2021. Tags: AI Engineering. "Sleep is not a disposable luxury - it's a non-negotiable biological necessity." -Matthew Walker. Questions: Why should you sleep for 8 hours? Why is sleep important for productivity? Does everyone need exactly eight hours?
- [The AI Monthly Top 3 - November 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-november-2021/) - Published Dec 05, 2021. Updated Dec 05, 2021. Tags: AI News and Analysis. The most interesting November's AI breakthroughs with video demos, short articles, code, and paper reference. Questions: Which papers appear in the November 2021 roundup? What does StyleCLIPDraw combine? What does SwinIR improve?
- [NVIDIA EditGAN: Image Editing with Full Control From Sketches](https://www.louisbouchard.ai/editgan/) - Published Dec 04, 2021. Updated Dec 05, 2021. Tags: Computer Vision, Generative AI, AI News and Analysis. Control any feature from quick drafts, and it will only edit what you want keeping the rest of the image the same! SOTA Image Editing from sketches model by. Questions: What does EditGAN allow users to edit? How does a segmentation map provide control? Can EditGAN preserve the rest of the image?
- [This AI makes blurry faces look 8 times sharper! SwinIR: Photo Upsampling](https://www.louisbouchard.ai/swinir/) - Published Nov 20, 2021. Updated Feb 03, 2025. Tags: Computer Vision. Transform your small 512-pixel large images into 4k with AI! Now you only need enough data, and you will have results like this (see the video!). Questions: What does SwinIR do? How can it make an image eight times larger? What does the Swin Transformer contribute?
- [Text-to-Drawing Synthesis With Artistic Control | CLIPDraw & StyleCLIPDraw](https://www.louisbouchard.ai/clipdraw/) - Published Nov 15, 2021. Updated Nov 16, 2021. Tags: Generative AI. Take a picture of the style you want to copy, enter the text, and the algorithm will generate a new picture out of it! Questions: What does CLIPDraw generate? What does StyleCLIPDraw add? Why is CLIP alone insufficient for style transfer?
- [The AI Monthly Top 3 - October 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-october-2021/) - Published Nov 01, 2021. Updated Nov 01, 2021. Tags: AI News and Analysis. The most interesting October's AI breakthroughs with video demos, short articles, code, and paper reference. Questions: Which papers appear in the October 2021 roundup? How does generative nowcasting predict rain? What is three-stem audio separation?
- [AI Synthesizes Smooth Videos from a Couple of Images!](https://www.louisbouchard.ai/ai-synthesizes-smooth-videos-from-a-couple-of-images/) - Published Oct 30, 2021. Updated Oct 30, 2021. Tags: Computer Vision. Let's construct 3D models from a couple of photos…. And thus, should we use this kind of model when humans are involved? Questions: How can a model create a smooth view from only a few photos? Is the result a conventionally recorded video? Why does image encoding matter for the result?
- [Isolate Voice, Music, and Sound Effects With AI](https://www.louisbouchard.ai/isolate-voice-music-and-sound-effects-with-ai/) - Published Oct 23, 2021. Updated Oct 24, 2021. Tags: Generative AI. If we take an audio clip with the music way too loud, you can simply turn up the speech and lower the music! Questions: What does audio source separation do? Can each separated track be adjusted independently? Why is creating training data for separation difficult?
- [Biologically-inspired Neural Networks for Self-Driving Cars](https://www.louisbouchard.ai/mit-biologically-inspired-neural-networks-for-self-driving-cars/) - Published Oct 17, 2021. Updated Oct 17, 2021. Tags: Large Language Models, AI Engineering. Biologically-inspired Neural Networks for Self-Driving Cars explains how language models behave, fail, improve. Questions: What is a neural circuit policy? Why explore smaller networks for autonomous driving? How are NCPs different from very deep supervised models?
- [DeepMind uses AI to Predict More Accurate Weather Forecasts](https://www.louisbouchard.ai/deepmind-rain-nowcasting/) - Published Oct 10, 2021. Updated Oct 10, 2021. Tags: AI News and Analysis. 50+ expert meteorologists assessed DeepMind's new model beating current nowcasting methods in 89% of situations for its accuracy and usefulness. Questions: What is precipitation nowcasting? How does DeepMind's model produce a rain forecast? Why use a generative model for weather?
- [The AI Monthly Top 3 — September 2021](https://www.louisbouchard.ai/ai-monthly-top-3-september-2021/) - Published Oct 02, 2021. Updated Oct 02, 2021. Tags: AI News and Analysis. The 3 most interesting AI papers this month with video demos, short articles, code, and paper reference. Questions: Which papers appear in the September 2021 AI roundup? What does StyleCLIP let a user change? What problem does TimeLens solve?
- [Generate Video Variations - No dataset or deep learning required!](https://www.louisbouchard.ai/vgpnn-generate-video-variations/) - Published Sep 25, 2021. Updated Sep 26, 2021. Tags: Generative AI, Learning AI. This model can do any video manipulation or video generation application you have in mind! Questions: What does VGPNN generate? Does the method need a separate training dataset? How does it extend image patch methods to video?
- [How to Make Slow Motion Videos With AI!](https://www.louisbouchard.ai/timelens/) - Published Sep 11, 2021. Updated Sep 25, 2021. Tags: Generative AI. TimeLens turns a few frames into slow-motion video. Here is how the model fills motion gaps and where generated video can break. Questions: What does TimeLens do? What is an event camera? Why combine frame and event cameras?
- [Manipulate Real Images With Text - An AI For Creative Artists! StyleCLIP Explained](https://www.louisbouchard.ai/styleclip/) - Published Sep 04, 2021. Updated Sep 25, 2021. Tags: Computer Vision, Generative AI. AI could generate images, then, using a lot of brainpower and trial and error, researchers could control the results following specific styles. Now. Questions: What does StyleCLIP do? How does text control the visual edit? Can StyleCLIP modify a real photograph?
- [The AI Monthly Top 3 — August 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-august-2021/) - Published Aug 29, 2021. Updated Sep 25, 2021. Tags: AI News and Analysis. The 3 most interesting AI papers of August 2021 with video demos, short articles, code, and paper reference. Questions: Which topics appear in the August 2021 roundup? What does SDEdit enable? How can sketches make GAN training more accessible?
- [Tesla's Autopilot Explained! Tesla AI Day in 10 Minutes](https://www.louisbouchard.ai/tesla-autopilot-explained-tesla-ai-day/) - Published Aug 21, 2021. Updated Sep 25, 2021. Tags: AI Engineering. Andrej Karpathy's talk on Tesla's autopilot explained clearly in under 10 minutes. Questions: What visual input does Tesla Autopilot use? Why are camera images rectified first? How are eight camera views combined?
- [Make GANs Training Easier for Everyone : Generate Images Following a Sketch](https://www.louisbouchard.ai/make-gans-training-easier/) - Published Aug 14, 2021. Updated Sep 25, 2021. Tags: Computer Vision, Generative AI. Control GANs outputs based on the simplest type of knowledge you could provide it: hand-drawn sketches. Questions: What does Sketch Your Own GAN allow? Why start from a pretrained GAN? What does fine-tuning change?
- [Image Synthesis and Editing from Sketches: SDEdit. No more tedious training needed!](https://www.louisbouchard.ai/image-synthesis-from-sketches/) - Published Aug 07, 2021. Updated Sep 25, 2021. Tags: AI Engineering, Computer Vision, Generative AI. Say goodbye to complex GAN and transformer architectures for image generation. This new method can generate new images from any user-based inputs. Questions: What does SDEdit do? Why add noise to the user's input? How does SDEdit preserve the original idea?
- [The AI Monthly Top 3 — July 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-july-2021/) - Published Jul 31, 2021. Updated Jul 31, 2021. Tags: AI News and Analysis. The 3 most interesting AI papers of July 2021 with video demos, short articles, code, and paper reference. Questions: Which projects appear in the July 2021 roundup? Why did GIRAFFE win attention at CVPR? What did Copilot demonstrate for programmers?
- [How Apple Photos Recognizes People in Private Photos Using Machine Learning](https://www.louisbouchard.ai/how-apple-photos-recognizes-people/) - Published Jul 31, 2021. Updated Sep 25, 2021. Tags: Computer Vision, Generative AI, Learning AI. Using multiple machine learning-based algorithms running privately on your device. Questions: How does Apple Photos recognize people? What is a face embedding? Why does on-device face recognition matter? Takeaways: Face recognition in photo apps usually depends on embeddings that group similar faces without needing a person to label every image first. On-device processing can improve privacy, but it does not remove the need to handle mistakes and sensitive identity data carefully.
- [OpenAI's New Code Generator: GitHub Copilot (and Codex)](https://www.louisbouchard.ai/github-copilot/) - Published Jul 24, 2021. Updated Sep 25, 2021. Tags: Computer Vision, Generative AI, AI News and Analysis. Find out how this AI Generates Code From Words. So if the data is randomly sampled from GitHub, how can you be sure it works and is well-written? Questions: What is GitHub Copilot? How can GitHub Copilot help developers? What should developers still review? Takeaways: GitHub Copilot is most useful as a coding assistant, not as a replacement for understanding the code you ship. The productivity gain comes from faster drafts, boilerplate, tests, and examples, but the review still belongs to the developer.
- [The Only AI Newsletter You Need](https://www.louisbouchard.ai/whats-ai-weekly-newsletter/) - Published Jul 22, 2021. Updated Jul 24, 2021. Tags: Learning AI, AI News and Analysis. There are already many machine learning newsletters, but this one is different and made for you! (contains videos, articles, code, demos…). Questions: What does the What's AI newsletter include? How is it different from a paper-title list? Who is the newsletter for?
- [⏩ How AI Generates New Images: GANs Put Simply](https://www.louisbouchard.ai/how-ai-generates-new-images/) - Published Jul 20, 2021. Updated Sep 01, 2021. Tags: Computer Vision, Generative AI. Understand how AI creates new images using Generative Adversarial Networks in 2 minutes. Questions: How does AI generate new images? Does AI copy images from the training set? Why do prompts change generated images? Takeaways: Image generation models learn patterns from training data, then combine those patterns into new images guided by prompts or conditions. The impressive part is not only the final image, but the representation the model learned about objects, styles, and composition.
- [⏩ How AI *Understand* Images in Simple Terms](https://www.louisbouchard.ai/how-ai-understand-images/) - Published Jul 19, 2021. Updated Sep 01, 2021. Tags: AI Engineering, Computer Vision. You can see CNNs as compressing the image, focusing on a specific feature of the image at every step. Questions: How does AI understand images? Does computer vision understand images like humans do? What are image features in AI? Takeaways: AI understands images by turning pixels into useful features, patterns, and representations for a task. The system does not see like a human. It learns statistical signals that can work well and still fail in strange ways.
- [How Machines Beat Humans at Everything](https://www.louisbouchard.ai/reinforcement-learning-icjai-21/) - Published Jul 17, 2021. Updated Jan 28, 2025. Tags: Learning AI. Discover how machines become super-human in most domains they attack using reinforcement learning. Questions: What is reinforcement learning? What are the main parts of an RL problem? Why does an agent explore randomly at first?
- [How AI Helps Spotting Wildfires](https://www.louisbouchard.ai/how-ai-helps-spotting-wildfires/) - Published Jul 14, 2021. Updated Sep 01, 2021. Tags: Learning AI. AI Spots Wildfires Faster Than Humans. Explained Simply. So, starting with this strong baseline they have. Questions: How can AI detect a wildfire? What training data does a smoke detector need? Why label smoke pixel by pixel?
- [CVPR 2021 Best Paper Award: GIRAFFE - Controllable Image Generation](https://www.louisbouchard.ai/cvpr-2021-best-paper/) - Published Jul 03, 2021. Updated Sep 25, 2021. Tags: AI Engineering, Computer Vision, Generative AI, AI News and Analysis. Using a modified GAN architecture, they can move objects in the image without affecting the background or the other objects! Questions: What does GIRAFFE generate? How does GIRAFFE improve controllability? Can one object move without changing the background?
- [The AI Monthly Top 3 — June 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-june-2021/) - Published Jun 30, 2021. Updated Jun 30, 2021. Tags: AI News and Analysis. The 3 most interesting AI papers of June 2021 with video demos, short articles, code, and paper reference. Questions: Which topics appear in the June 2021 roundup? What makes DefakeHop practical? What does Barbershop edit?
- [Create Realistic Animated Looping Videos from Pictures](https://www.louisbouchard.ai/animate-pictures/) - Published Jun 26, 2021. Updated Sep 01, 2021. Tags: Computer Vision, Learning AI. a) Input Image, b) Eulerian field, and c) movement needed in the frames. Holynski, Aleksander, et al., 2021. Questions: How can one picture become a looping animation? What is an Eulerian flow field in this method? Which images work naturally with this approach?
- [How to Read More Research Papers?](https://www.louisbouchard.ai/research-papers/) - Published Jun 23, 2021. Updated Sep 25, 2021. Tags: Learning AI, AI News and Analysis. A practical guide to reading more AI research papers without getting buried, with habits for choosing, skimming, and revisiting papers. Questions: How to Read More Research Papers? How should beginners read AI research papers? What should you look for in an AI paper? Takeaways: Reading AI papers is easier when you separate the claim, the method, the experiment, and the actual evidence. You do not need to understand every equation on the first pass. Start with the problem, result, assumptions, and limitations.
- [Translate or Edit Text from Images Emulating the Style: TextStyleBrush](https://www.louisbouchard.ai/textstylebrush/) - Published Jun 19, 2021. Updated Sep 01, 2021. Tags: Computer Vision. This new Facebook AI model can translate or edit every text in the image in your own language, following the same style! Questions: What does TextStyleBrush do? How many style examples does it need? Can it write a different language?
- [Barbershop: Try Different Hairstyles and Hair Colors from Pictures (GANs)](https://www.louisbouchard.ai/barbershop/) - Published Jun 12, 2021. Updated Sep 25, 2021. Tags: Computer Vision, Generative AI. This AI can transfer your hair to see how it would look like before committing to the change. Questions: What can Barbershop change in a portrait? Why is hairstyle transfer harder than copying pixels? How does lighting affect a transferred hairstyle?
- [How to Spot a Deep Fake. Breakthrough US Army technology (2021)](https://www.louisbouchard.ai/spot-deepfakes/) - Published Jun 05, 2021. Updated Sep 01, 2021. Tags: Generative AI. How deepfake detection works, why artifacts matter, and what a US Army research project showed about spotting generated faces. Questions: How does a deepfake detector identify generated faces? Why can a model detect artifacts people miss? Why was the US Army detector notable?
- [The AI Monthly Top 3 — May 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-may-2021/) - Published May 31, 2021. Updated May 31, 2021. Tags: AI News and Analysis. a curated list of the latest breakthroughs in AI and Data Science in May 2021 with a clear video explanation, link to a more in-depth article. Questions: Which papers appear in the May 2021 roundup? What does Total Relighting solve? What does LASR reconstruct?
- [High-Resolution Photorealistic Image Translation in Real-Time](https://www.louisbouchard.ai/4k-image-translation-in-real-time/) - Published May 29, 2021. Updated Sep 01, 2021. Tags: Computer Vision, Learning AI. Apply any style to your 4K image in real-time using this new machine learning-based approach! Questions: What is real-time 4K image-to-image translation? Why does the method split low- and high-frequency information? How does this split make 4K translation faster?
- [Is AI The Future Of Video Game Design? Enhancing Photorealism Enhancement](https://www.louisbouchard.ai/the-future-of-video-game/) - Published May 22, 2021. Updated May 22, 2021. Tags: AI Engineering. This AI can be applied live to the video game and transform every frame to look much more natural. Questions: What is photorealism enhancement for games? Why use engine buffers instead of only the screenshot? Can the model run on live gameplay?
- [Articulated 3D Reconstruction from Videos](https://www.louisbouchard.ai/3d-reconstruction-from-videos/) - Published May 15, 2021. Updated May 15, 2021. Tags: Computer Vision, Generative AI. Generate 3D models of humans or animals moving from only a short video as input. Questions: What does articulated 3D reconstruction produce? What input does the reconstruction method require? Why does the method calculate optical flow?
- [Realistic Lighting on Different Backgrounds](https://www.louisbouchard.ai/backgrounds-with-lighting/) - Published May 08, 2021. Updated May 09, 2021. Tags: AI Engineering. Properly relight any portrait based on the lighting of the new background you add. Questions: Why does a replaced background often look unrealistic? What is human matting? What does the relighting module change?
- [Infinite Nature: Fly into an image and explore the landscape](https://www.louisbouchard.ai/infinite-nature/) - Published May 01, 2021. Updated May 06, 2021. Tags: Computer Vision, Generative AI. The next step for view synthesis: Perpetual View Generation, where the goal is to take an image to fly into it and explore the landscape! Questions: What is perpetual view generation? Why does the method need a disparity map? What happens after the viewpoint moves?
- [The AI Monthly Top 3 — April 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-april-2021/) - Published May 01, 2021. Updated May 06, 2021. Tags: AI News and Analysis. a curated list of the latest breakthroughs in AI and Data Science in April 2021 with a clear video explanation, link to a more in-depth article. Questions: What is included in the April 2021 AI roundup? What does GANverse3D contribute? Why was computer-vision progress reviewed that month?
- [What is the state of AI in computer vision?](https://www.louisbouchard.ai/ai-in-computer-vision/) - Published Apr 25, 2021. Updated Sep 01, 2021. Tags: Computer Vision. I will openly share everything about deep nets for vision applications, their successes, and the limitations we have to address. Questions: What is the state of AI in computer vision? What is AI in computer vision? Why does data matter so much in computer vision? Takeaways: Computer vision is useful when the model solves a clear perception task, not only when the demo image looks impressive. Data quality, labels, camera conditions, and edge cases usually decide whether a vision system works in the real world.
- [The AI Monthly Top 3 — March 2021](https://www.louisbouchard.ai/ai-monthly-march-2021/) - Published Apr 25, 2021. Updated Apr 25, 2021. Tags: AI News and Analysis. The 3 most interesting AI papers this month, March 2021, with video demos, short articles, code, and paper reference. Questions: What is included in the March 2021 AI paper roundup? How did the brain-computer interface study model attractiveness? What question did the AI dating-profile project explore?
- [GrabCut NoteBook implementation for Image Matting](https://www.louisbouchard.ai/grabcut-notebook-implementation/) - Published Apr 21, 2021. Updated May 06, 2021. Tags: Computer Vision, AI News and Analysis. A fun introduction project for computer vision and python on Google Colab. iterative-grabcut: Using a simple rectangle and a brush removing the background of. Questions: What does the GrabCut notebook do? Is GrabCut a good beginner computer-vision project? What does the initial rectangle provide?
- [Create 3D Models from Images! GANverse3D & NVIDIA Omniverse](https://www.louisbouchard.ai/ganverse3d/) - Published Apr 18, 2021. Updated Sep 01, 2021. Tags: Computer Vision, Generative AI, AI News and Analysis. This promising model called GANverse3D only needs an image to create a 3D figure that can be customized and animated! Questions: What does GANverse3D create? Why is single-image 3D reconstruction difficult? Which object categories did the application support?
- [An Amputee with an AI-Powered Hand! 🦾](https://www.louisbouchard.ai/an-amputee-with-an-ai-powered-hand/) - Published Apr 11, 2021. Updated Apr 27, 2021. Tags: AI Engineering. With this AI-powered nerve interface, the amputee can control a neuroprosthetic hand with life-like dexterity and intuitiveness. Questions: What does the AI-powered prosthetic hand allow its user to do? How does the prosthesis receive movement commands? Why is model accuracy important in a neuroprosthesis?
- [How can Transformers be used in Computer Vision?](https://www.louisbouchard.ai/will-transformers-replace-cnns-for-vision/) - Published Apr 08, 2021. Updated Sep 25, 2021. Tags: AI Engineering, Computer Vision, AI News and Analysis. In less than 5 minutes, you will know how the transformer architecture can be applied to computer vision with a new paper called the Swin Transformer. Questions: How can Transformers be used in Computer Vision? What is a vision transformer? Will transformers replace CNNs in computer vision? Takeaways: Vision transformers changed computer vision by applying attention to image patches, but that does not make CNNs obsolete overnight. Transformers can work extremely well with enough data and compute, while CNNs can still be efficient and strong for many practical tasks.
- [The AI Monthly Top 3 — February 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-february-2021/) - Published Mar 03, 2021. Updated Apr 25, 2021. Tags: AI News and Analysis. Odei Garcia-Garin et al. from the University of Barcelona have developed a deep learning-based algorithm able to detect and quantify floating garbage from. Questions: Which papers appear in the February 2021 roundup? How can AI help monitor ocean plastic? What does ShaRF reconstruct?
- [OpenAI’s DALL·E: Text-to-Image Generation Explained](https://www.louisbouchard.ai/openais-dall-e-text-to-image-generation-explained/) - Published Feb 27, 2021. Updated Apr 25, 2021. Tags: Computer Vision, Generative AI, AI News and Analysis. OpenAI just released the paper explaining how DALL-E works! It is called “Zero-Shot Text-to-Image Generation”. Questions: What is DALL-E? Why was DALL-E important? How does text-to-image generation work? Takeaways: DALL-E showed how language and image generation can connect, turning text prompts into visual concepts. The model is useful because it can combine ideas, objects, and styles, but the prompt is still only partial control.
- [The AI Monthly Top 3 — January 2021](https://www.louisbouchard.ai/the-ai-monthly-top-3-january-2021/) - Published Feb 04, 2021. Updated Apr 25, 2021. Tags: AI News and Analysis. What happened in the AI research in January 2021. The 3 most interesting papers! Questions: Which papers defined the January 2021 roundup? What did DALL-E demonstrate? What problem did VOGUE address?
- [Third Wave of AI | Thinking Fast and Slow](https://www.louisbouchard.ai/third-wave-of-ai-thinking-fast-and-slow/) - Published Jan 20, 2021. Updated Apr 25, 2021. Tags: AI Engineering. Drawing inspiration from Human Capabilities Towards a more general and trustworthy AI & 10 Questions for the AI Research Community. Questions: What does fast and slow thinking mean for AI? Why is this framing useful for AI builders? Where do neural networks fit in this idea? Takeaways: The fast-and-slow framing is useful because AI systems need both pattern recognition and more deliberate reasoning or checking. Fast predictions are powerful, but they can be brittle when the task needs planning, uncertainty, memory, or explicit constraints.
- [2020: A Year Full of Amazing AI Papers - A Review](https://www.louisbouchard.ai/2020-a-year-full-of-amazing-ai-papers-a-review/) - Published Jan 05, 2021. Updated Apr 25, 2021. Tags: AI News and Analysis. A curated list of the latest breakthroughs in AI in 2020 by release date with a clear video explanation, link to a more in-depth article, and code. Questions: What does this 2020 AI research review include? What did DeepFaceDrawing make possible? Why was PULSE an interesting image-generation result?
- [What is it to be a Data Scientist - An Interview](https://www.louisbouchard.ai/interviewing-a-data-scientist/) - Published Dec 20, 2020. Updated Apr 25, 2021. Tags: AI Engineering. A professional data scientist answers the 12 most asked questions. Clarify and debunks the myths you've built surrounding this field! Questions: What does a data scientist do? What skills matter most for data science? How can beginners build a data science portfolio? Takeaways: A strong data scientist is not only a model builder. The job also needs problem framing, communication, and business judgment. Projects matter because they show how someone handles messy data, unclear goals, tradeoffs, and explaining results to other people.
- [Train Your GAN With 1/10th of the Data! NVIDIA ADA Explained](https://www.louisbouchard.ai/nvidia-ada/) - Published Dec 07, 2020. Updated May 05, 2021. Tags: Computer Vision, Generative AI, AI News and Analysis. With this new training method developed by NVIDIA, you can train a powerful generative model with one-tenth of the images! Making possible many applications. Questions: What is NVIDIA's adaptive discriminator augmentation? Why do GANs overfit with limited images? How does augmentation help the discriminator?
- [Remove any Background Without Green Screens](https://www.louisbouchard.ai/remove-background/) - Published Dec 03, 2020. Updated Sep 25, 2021. Tags: Computer Vision. This new background removal technique can extract a person from a single input image, without the need for a green screen in real-time! Questions: What is real-time image matting? Why is matting harder than a hard segmentation mask? How does GrabCut remove a background?
- [State-of-the-Art Convolutional Neural Networks Explained - DenseNet](https://www.louisbouchard.ai/densenet-explained/) - Published Dec 02, 2020. Updated Sep 25, 2021. Tags: Computer Vision. Facial recognition, targeted ads, and image recognition all have one thing in common. They are all done using the best Convolutional Neural Networks. Questions: What is a convolutional neural network? What makes DenseNet different from a standard CNN? Why reuse features across layers?
- [DeOldify: Colorize your Black & White Photos with AI](https://www.louisbouchard.ai/deoldify/) - Published Nov 08, 2020. Updated Sep 25, 2021. Tags: AI Engineering. This method is called DeOldify and works on pretty much any picture. If you don’t believe me, you can even try it yourself for free. Questions: What does DeOldify do? What is NoGAN training? Why pretrain the generator first?
- [Transfer clothes between photos using AI](https://www.louisbouchard.ai/transfer-clothes-between-photos/) - Published Nov 03, 2020. Updated Sep 01, 2021. Tags: AI Engineering. This artificial intelligence model transfers clothes between photos using only one picture of yourself and one picture of the clothing! This is called garment. Questions: What is garment transfer? Which inputs does the model need? How does the model account for body pose?
- [Toonify: Transform Faces into Disney animated movie Characters](https://www.louisbouchard.ai/toonify/) - Published Oct 25, 2020. Updated Nov 15, 2023. Tags: AI Engineering. This AI can transform any of your pictures into an accurate representation with a Disney animated movie character style! Questions: What does Toonify do? Which base model powers Toonify? How is a real face converted into the new style?
- [Create and Understand DeepFakes in 5 minutes](https://www.louisbouchard.ai/deepfakes-in-5-minutes/) - Published Oct 13, 2020. Updated Sep 25, 2021. Tags: AI Engineering. Understand how deep fakes work and create your own! You should definitely give it a shot if deepfakes interest you! Questions: What is a deepfake? How do deepfakes work? Why are deepfakes risky? Takeaways: Deepfakes matter because image and video generation can now change trust, identity, and evidence, not only entertainment. The technical idea is pattern transfer: models learn enough about faces, motion, and style to synthesize convincing media.
- [Can GPT-3 Really Help You and Your Company?](https://www.louisbouchard.ai/can-gpt-3-really-help-you/) - Published Sep 13, 2020. Updated Sep 25, 2021. Tags: Large Language Models. If you have ever wondered either what's GPT-3 and how can it be useful to you or your company, this is the article you were looking for. Questions: Can GPT-3 Really Help You and Your Company? Can GPT-3 help a company? What GPT-3 use cases are easiest to test first? Takeaways: GPT-3 is useful when it is attached to a specific language workflow, not when it is treated as general magic. The first tests should measure quality, cost, privacy, hallucinations, and where human review still belongs.
- [What is the YOLO algorithm? Introduction to Real-Time Object Detection](https://www.louisbouchard.ai/you-only-look-once/) - Published May 28, 2020. Updated Sep 01, 2021. Tags: AI Engineering. YOLO, Also Known as You Only Look Once is one of the most powerful real-time object detector algorithms. It is called that way because…. Questions: What is YOLO in computer vision? Why is YOLO popular? What is object detection? Takeaways: YOLO is useful because it frames object detection as one fast prediction over the image, which made real-time detection practical. Speed is the headline, but the real test is whether detections stay reliable across lighting, scale, occlusion, and camera changes.