Reasoning Models Don't Always Say What They Think

Reasoning Models Don't Always Say What They Think

Why it matters

Measured that CoTs mention hints they actually used mostly less than 20% of the time

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2505.05410

Added in Week 35, 2026 · See that week

Nearby on the same axis

Grok inserts South African "white genocide" into unrelated answers for hoursPressPreparedness Framework v2OfficialAnthropic activates ASL-3 for the first time (Claude Opus 4)OfficialOn the Biology of a Large Language ModelOfficial