The Genius with a goldfish memory
That’s how I explained AI to people not yet working with it. That was my first thought when I first encountered LLMs when GPT-3.5 came out.
The idea that every conversation is a fresh start, where you get to discuss and exchange with a sharp mind, except that they don’t really know it nor remember it.
With the latest SOTA models, this is even more striking, what used to be a ~10 year-old Genius back in GPT-3.5 times, is now a Fields medalist / Nobel prize multi PhD 30ish genius that doesn’t really know it until you remind them, or ask them specific questions.
I've found this personification useful because it has helped me bring AI into people’s workflow and mentor people to use AI better.
This framing also helps to realize that it’s all about the context: reminding the AI at the right time, about the right things, as well as helping them navigate work by putting virtual post-it notes everywhere: “this is where you can find X, this is where you can find Y”.
And what’s beautiful is that you can take advantage of the bad memory of the genius: adversarial reviews of work where you hand over work done by someone else (an AI) and ask this freshly minted genius their opinion: fresh on the ground, with no attachment to the existing work or knowledge of its history, you’ll be sure to gain interesting insights.
Or prompting “you’re an award winning motion designer that despises the use of AI and spots it right away” to tweak the video you made with Opus 5.5 that is a tad too AI for your taste.
With this intelligence on tap, where any arcane or esoteric knowledge can be retrieved by just reminding the genius (that may do some research on their own), it becomes evident that being well-travelled, well-cultured and well-spoken becomes a significant advantage.
Sometimes a single word (a concept, a culture reference) is enough to invoke an entire body of knowledge. I call these context shortcuts.
Mention poka-yoke when designing a system and you’ve introduced an entire philosophy of mistake-proofing.
Ask the genius to behave like Dwight from The Office and you now have a pedantic colleague that will nitpick on everything you say.
Ask for a Dieter Rams-ish redesign of your office, UI or house to invoke functional simplicity.
Audit your work by asking the genius to act as a red team, and you have a reviewer actively hunting for weaknesses in your system.
Ask the genius to respect Chesterton's fence, forcing it to understand why something exists before proposing changes.
And so there’s this really interesting paradox where the smarter the AI becomes, the bigger the advantage of being a polymath / intellectually curious generalist becomes.
Exciting times. The answers are increasingly within reach: you just need to know how to ask. And of course what to do with the answers.