LLMs' "simulated reasoning" abilities are a "brittle mirage," researchers find

5 merksittich 2 8/11/2025, 9:45:58 PM arstechnica.com ↗

Comments (2)

merksittich · 4h ago
Preprint discussed in the article:

https://arxiv.org/pdf/2508.01191

verdverm · 2h ago
fyi, this is about Chain-of-thought, not <think>, is that still being used?

haven't read the paper closely enough to comment on the methods