Unverified50% confidenceFactExact time
In Willison's experiment, a single answer consumed 46 minutes and 43,506 reasoning tokens.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Related Entities
Related Claims
UnverifiedThe format line directly determines how often the model shows its reasoning process: "Answer with the number only" led to 37% of responses showing reasoning, the plain version 27%, and the florid version only 8%.69% similarUnverifiedTests on five models show that whole-question accuracy on the Reproduce task ranges from 49.21% to 68.25%.66% similarUnverifiedAfter the cleanup, 42% of the bot's replies cited at least one saved answer, drawn from more than 20 different answers.66% similarUnverifiedAn Anthropic randomized trial had 52 engineers (mostly junior) learn an unfamiliar Python library; the AI-assisted group averaged 50% on the comprehension quiz, while the handwritten-code group averaged 67%, with the largest gap on debugging-type questions; the speed advantage brought by AI was about 2 minutes and was not statistically significant.66% similarUnverifiedIn an ablation experiment, after replacing the model's latent thinking segments with segments from unrelated questions, accuracy dropped by 45%–51%, indicating that latent thinking carries substantive reasoning.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135400API
curl https://kongchang.com/api/v1/knowledge/claims/1135400MCP
get_claim(id=1135400)