I'm partnering with Mike Knoop to launch ARC Prize: a $1,000,000 competition to create an AI that can adapt to novelty and solve simple reasoning problems. Let's get back on track towards AGI. Website: ARC Prize on Kaggle:

I published the ARC benchmark over 4 years ago. It was intended to be a measure of how close we are to creating AI that can reason on its own – not just apply memorized patterns.

ARC tasks are easy for humans. They aren't complex. They don't require specialized knowledge – a child can solve them. But modern AI struggles with them. Because they have one very important property: they're designed to be resistant to memorization.

Here's the thing: what current AI (e.g. LLMs) is really doing is memorizing millions of patterns seen in human-generated data, and reapplying them on new inputs. That works great when you're dealing with a well-known problem – until you introduce any amount of novelty.

But the nature of intelligence is precisely to adapt to things you don't expect. To figure out what to do when you don't have a solution already memorized. If your AI can't adapt to novelty, it will never be able to deal with the variability and fluidity of the real world.

And that's why LLMs aren't on the path to AGI. They cannot reason – they recite. They by-pass the need for intelligence by leveraging memorization instead – on a scale that boggles the mind. And ARC is like a flashing red light reminding you that we're missing something.

We want ARC Prize to be, first of all, a barometer of how close or how far we are to developing general AI, and second, a source of inspiration for AI researchers. We want people to look at ARC and ask, what would it take to solve these problems?

We want to nudge you into looking at questions you're not currently asking, and try ideas you would not normally pursue. We want to set the field back on track towards AGI. Good luck... Many thanks to all involved in the project – in particular our friends from Kaggle, Lab42, and InfiniteMonkey.

@mikeknoop @kaggle ARC solvers: Before submitting to the ARC Prize, try out ConceptARC, a set of easier, concept-based tasks in the ARC domain. Paper: Corpus: Let me know how your program does!

@mikeknoop @kaggle So happy to see this, François! :D

@mikeknoop @kaggle You can try out a version of the challenge here: