What counts as illegible reasoning?
SummaryIllegible reasoning in LLMs has been observed in OpenAI models, and understanding this behavior would be beneficial for AI safety research. This post describes challenges with reproducing this behavior in open models and limitations of LLM-as-ju…