Unverified50% confidenceFactExact time
Existing evaluations almost always look at only one metric: whether the log-probability of the target token decreases after intervention.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Related Entities
Related Claims
UnverifiedShannon pointed out that if an event has n equally probable possible outcomes, the minimum amount of information needed to determine the specific outcome is log₂(n) bits.69% similarUnverifiedApplying a logit bias of -2 reduces the probability of the target tokens to approximately 1/7 of the original (e^-2 ≈ 0.135).69% similarUnverifiedAblation experiments and behavioral analysis show that the tokens saved by DoubtLess mainly come from after the first appearance of the correct answer.69% similarUnverifiedIn evaluation regression detection, only if the candidate's lower bound is higher than the baseline's lower bound (or the baseline point estimate) is there statistical grounds to say it has not gotten worse; otherwise it can only be judged inconclusive.68% similarUnverifiedIf the highest score after reranking is below a threshold validated against a test set, the model should be told that no relevant content was found.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1133778API
curl https://kongchang.com/api/v1/knowledge/claims/1133778MCP
get_claim(id=1133778)