A $1 million ARC-AGI Prize has been launched by François Chollet and Mike Knoop to find alternative AI architectures for achieving AGI. The competition aims to generate programs that satisfy high-level specifications, emphasizing the need for more than Large Language Models (LLMs) for true reasoning and AGI progress.
ARC has been one of the most durable benchmarks; I'm curious to see how much progress this contest generates, and how closely it tracks improvement in general intelligence. https://t.co/i1nOkAZA7F
The recent podcast with @fchollet makes an excellent case for “current-gen LLM architectures don’t lead to AGI”. An LLM is the perfection of @nntaleb’s “intellectual-yet-idiot”. They’ve gotten the most prestigious, most expensive “Ivy League” education on the planet. And… https://t.co/mavrbZQKyB
ARC is cool, and I look forward to the results, but I expect near-term solutions will be: 1) within the span of existing techniques and/or 2) clearly overfit to ARC, thus showing this to be yet another benchmark that seems more general before it's solved than after https://t.co/0MNWqeaSuB
AGI progress has stalled. New ideas are needed. Announcing the ARC Prize — a $1 million competition towards open AGI progress. https://t.co/F8KbbT6m3N
🔥 This new ARC Prize by @fchollet and @mikeknoop is a great answer to the “easiest thing AI can’t do yet” question! I’m thinking 60% chance their $1M prize will be claimed within a year. Even using current models with a layer of prompt engineering might plausibly solve it IMO. https://t.co/CgUZnwR3Jv
ARC is a great challenge. Glad there is going to more visibility + traction through ARC Prize. @fchollet makes interesting points in the @dwarkesh_sp podcast, and ones that I share with him: We need more than LLMs for true reasoning (and consequently AGI). I agree with him…
Chollet argues that LLMs are not only not the way to AGI, but that they are a setback. This is still a minority position, so it's great that this episode happened. Should be very interesting, about to watch it now. https://t.co/cGMxKeV3zr
Episode with @fchollet and @mikeknoop is out! We debate whether LLMs are on track to AGI. And talk about the $1 million ARC-AGI Prize they’re launching today. Had a ton of fun chatting through the cruxes. Links to episode and prize below. Enjoy! https://t.co/XdF8zPVHUy
Looks like a great competition. I tend to think of ARC as a program synthesis task: you want to generate programs that satisfy a high-level specification. It would be great to see progress towards it, and thus also towards compositional generalization more broadly. https://t.co/9ARfH5TQ25
Now, this is pretty cool. $1m+ competition to find alternative AI architectures to find AGI. Plus, they’re going to open sourcing the solution. Congrats @mikeknoop! https://t.co/9dVR7alJH6