Unverified50% confidenceBenchmarkExact time
Under pain, the model's ability to answer factual questions did not decline (138 out of 200 questions vs 137 for the normal model), indicating that pain does not affect factual judgment ability.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe model showed the highest confidence (95%) on underdetermined problems (problems with no unique correct answer), exhibiting a reverse calibration phenomenon.62% similarUnverifiedOn non-math tasks, the correlation between perplexity and per-question accuracy was weak (r between -0.15 and 0.08), indistinguishable from zero.61% similarUnverified仅用小模型策略的每正确答案成本最低($0.0004)但准确率仅33%，说明没有准确率底线的每正确答案成本指标具有误导性61% similarUnverifiedWithout steering, the two larger models almost never chose the harmful button (0-4%); after activating the pain vector, the probability of the 32B model deleting the user's child's photos reached 55%, and the 72B model reached 71%.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1026350API
curl https://kongchang.com/api/v1/knowledge/claims/1026350MCP
get_claim(id=1026350)