METR says frontier AI companies confirmed their models cannot perform long serial reasoning without explicit chains of thought
METR cannot verify these limits for non-participating companies
——0——
@ChrisPainterYup @a_karvonen I don't remember the rumor precisely but it seemed alluding to something UT like (https://arxiv.org/abs/1807.03819) which would both might allow for test-time scaling/train-time adaptive compute at token level, but wouldn't allow for serial reasoning without CoT, given recurrence != CoT.
Chris Painter@ChrisPainterYup@willdepue @a_karvonen I'm not that I understand you, so I'm not sure if this addresses your question, but see:
4:39 AM · May 26, 2026 · 217 Views
4:54 AM · May 26, 2026 · 202 Views