There is a new secret method, a new skill, that will make you 10 times more effective. It is AI's unique language: prompting.
Here are a few key insights from my interview with Sander Schulhoff, creator of learnprompting.org (the most comprehensive guide on how to talk with AI (prompting)), which is available on the What’s AI podcast available on YouTube, Spotify, and Apple podcasts.
So, what is prompting?
Prompting is talking with AIs. Any written text (instruction) is a good way of thinking about it. “draw me an image of a cow” or “write an email to my boss about me coming in sick to work today” are prompts.
It can also be non-text content like images.
Sander, the creator of the learnprompting.org platform, defines prompting as being “artificially socially intelligent”. “To communicate with AIs, you'll need to speak to them similarly to how humans communicate with humans (social intelligence) but involving AI. It becomes something new like artificial social intelligence.”
Why learn prompting?
The models understand things differently than humans.
You may ask it to do something obvious to you, but it does the opposite.
Like communication skills, better prompting allows you to have what you want. Have finer control over what the model outputs.
Could you give me an example of “a better prompting”?
You could ask it to write an email, and it writes a boring and blend one.
So if instead of saying write me an email to my boss you say write me a detailed and funny email to my boss, make at least 2 jokes Throughout. You're going to end up with a better email.
How to improve our prompting skills?
This can be done through strategies like:
- zero-shot chain of thought; asking the model to think through its steps.
- Few-shot prompting; showing the model an examples of what you expect it to do.
More details on the free learnprompting.org platform!
A concrete example for me is to use ChatGPT to rephrase my scripts (as a non-native English speaker). It helps me produce better videos using it as an editor vs. an end-to-end writer. (see image below)
Such AIs are not "ready" to do so, nor safe (high plagiarism concerns due to their training data).