Just made a repository with golfed solutions to ARC-AGI-1-train for the NeurIPS 2025 Google Code Golf Championship as well as some golfing tricks that I learned:

There's been significant recent progress in the NanoGPT speedrun. Highly recommend this post by @classiclarryd https://t.co/QNoI4wVAJg

Excited to advance our lead and SoTA score on ARC-AGI-2 (@arcprize) by 3 points to 15.28. @DriesSmit1 @MohamedOsmanML @bayesilicon @GregKamradt @tufalabs https://t.co/GCJvSlO0hm

🚨 NEW PAPER DROP! Wouldn't it be nice if LLMs could spot and correct their own mistakes? And what if we could do so directly from pre-training, without any SFT or RL? We present a new class of discrete diffusion models, called GIDD, that are able to do just that: 🧵1/12

Happy to announce we outperformed @OpenAI o1 with a 7B model :) We released two self-improvement methods for verifiable domains in our preliminary paper -->

Today, MindsAI (@MindsAI_Jack @MohamedOsmanML @bayesilicon) becomes part of @tufalabs First assignment: complete the @arcprize challenge

Great presentation on some unique TTT ideas and experiments by Jonas Hübotter @tufalabs. https://t.co/kyZOcoJ7v3

Consulting my heart... Ok, looks like you haven't. But whenever you have a SotA (or close) solution built on top of the OpenAI API we're more than happy to verify it and add it to the public ARC Prize leaderboard. Anything using less than $10k worth of API calls is eligible.

ARC prize 2024 🥈place paper by the ARChitects who scored 53.5 (56.5): https://t.co/2rrtqJgRNL - Transformers/LLMs are for ARC what ConvNets were for Imagenet - strong base model, TTT, specialized datasets (e.g. @bayesilicon’s re-arc) + novel: DFS sampling with LLM critique

very excited to win guys, it's been such a blast! let's goo

Have been working on my 2nd synthetic ARC riddle generator (agent: ideation -> prog generation). Got >1k diverse generator+solver pairs as PoC so far. Some nice examples:

I finally got to meet @fchollet in person recently to interview him about @arcprize, intelligence vs memorization, human cognitive development, learning abstractions, limits of pattern recognition and consciousness development. These are the best bits. Full show released tomorrow

We got upto 55.5% on the @arcprize leaderboard today! Progress towards the 60.2 % milestone of median human performance reported by https://t.co/lH4lIViiLh is not slowing down. @MindsAI_Jack @bayesilicon

New ARC-AGI paper @arcprize w/ fantastic collaborators @xu3kev @HuLillian39250 @ZennaTavares @evanthebouncy @BasisOrg For few-shot learning: better to construct a symbolic hypothesis/program, or have a neural net do it all, ala in-context learning? https://t.co/zcmxoQzv92

New SoTA on ARC-AGI. Nothing like the synergy of an awesome team (@bayesilicon @MohamedOsmanML). From 53 to 54.5 today. Onward and upward! 🚀 @arcprize @mikeknoop @fchollet @bryanlanders @GregKamradt @MLStreetTalk First. #kaggle -

yes

just achieved a score of 53% on the @arcprize - what a feeling! @MindsAI_Jack @MohamedOsmanML lets gooo!

We trained a Vision Transformer to solve ONE single task from @fchollet and @mikeknoop’s @arcprize. Unexpectedly, it failed to produce the test output, even when using 1 MILLION examples! Why is this the case? 🤔

when ARC merch

New ARC-AGI high score! 48% (Prize goal: 85%) Congratulations, MindsAI!