1
ARC-AGI without pretraining
We solve 20% of ARC-AGI-1 with no pretraining by minimizing the description length during inference time, with the only data involved being the target puzzle itself.
I am a PhD student in the Machine Learning Department at Carnegie Mellon University. I recently completed my Master's under Max Tegmark, researching mechanistic interpretability.
Visit iliao2345.github.io → On Hacker News Mar 2025 – Mar 2025
We solve 20% of ARC-AGI-1 with no pretraining by minimizing the description length during inference time, with the only data involved being the target puzzle itself.