Another short critique of the Anthropic "Hot Mess" paper

A new critique argues Anthropic's viral 'Hot Mess' paper—which claimed AI models become more incoherent as they use more reasoning tokens—reaches conclusions not supported by its experiments. The analysis suggests task difficulty and inherent question properties could explain the results, not model behavior. Despite controlling for difficulty with per-question analysis, the critique claims the paper's findings remain 'underdetermined' and could reflect dataset artifacts rather than actual AI misalignment patterns.