scalingfactbearishCurrent base LLMs still do not perform well on ARC 1 and can't even reliably do simple math operationsFrancois Chollet08 Aug 2026https://x.com/fchollet/status/2085416724266983682