Welcome to Last Week in AI, a post I publish every Friday to share a couple things I’ve discovered in the world of AI last week. I spend way too much time in Discord, on Twitter and browsing reddit so you don’t have to!
If you have a tip or something you think should be included in next week’s post, send an email to keanan@floorboardai.com with more info.
This week we’ve got an example of how not to frame things when you’re talking to AI and an example of a AI-enabled word/crafting game.
Let’s dive in!
Don’t use negative framing with AI
When I was growing up, my mom told me about how she learned in flight attendant training to give passengers positive instructions on what to do instead of negative “don’t” instructions on what not to do.
For example, instead of saying “don’t bring your bags with you” tell people to “move as quickly as you can down the aisle, off the slide and away from the airplane”. The idea behind this was that, in a crisis, people hearing the “don’t” language just heard the thing you were telling them not to do, and would do that thing based on what they heard.
It turns out, this idea seems to apply to AI as well, as many examples this week have shown. A prompt of “show me a room that doesn’t include an elephant” definitely includes an elephant, for example:
A cool crafting game
If you don’t know who Neal Agarwal is, you should definitely check out his site, neal.fun. You can try a ton of his digital experiments including Space Elevator (where you can scroll up to see the heights of various natural phenomena), Design the Next iPhone (where you can add as many features to a virtual iPhone as you want, to the point of absurdity), Spend Bill Gates’ Money (one of his more popular projects, where you get a sense of just how much money Mr. Gates has), and more.
He is truly the master of creating fun and interesting Internet projects and you can lose hours just kicking around his site. The exciting news this week is that he’s just launched something new.
This game, called Infinite Craft, starts off simple, giving you “Water”, “Fire”, “Wind” and “Earth” to drag onto a canvas. When you drop one element on top of another, it combines them. You can them combine those combinations with other base elements, or with other combinations that you’ve already created.
Behind the scenes, a large language model (LLM) is deciding what your combinations create! This is a great example of a game that would have been insanely difficult pre-LLM, but now that we have LLMs, is something relatively easy to build and deploy.
See you next week!
If you’ve made it this far and enjoyed “Last Week in AI”, please drop your email down below so next week’s edition goes straight to your inbox.