Bookmark · 12 August 2025, 19:30 LLMs’ “simulated reasoning” abilities are a “brittle mirage,” researchers find - Ars Technica ml-accuracy