Unverified50% confidenceDecision RuleExact time
logprob thresholds need to be calibrated for the specific model and task distribution; directly applying a fixed threshold often works poorly.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn experiments, the anomaly score (the distance between the current latent state and the expected latent state) can serve as a trigger for maintenance intervention, and when it exceeds a threshold for several consecutive time steps the strategic agent requests a maintenance operation.68% similarUnverifiedOnTrack's limitation is that the quality of the reference trajectories directly determines the monitoring ceiling; if the historical successful runs themselves have problems, the comparison baseline is unreliable.67% similarUnverifiedIn lossy speculative decoding scenarios, reporting only the number of accepted tokens per round is insufficient to measure decoding speed; at minimum, completion length and number of verification rounds must be reported together, and judgments should be made in combination with actual elapsed time and task accuracy.66% similarUnverifiedThe "drift threshold hypothesis" predicts the existence of a complementary failure clock whose sample count remains approximately invariant under the square-root rule.66% similarUnverifiedThe effectiveness of threshold functions is highly dependent on the calibration quality of the underlying model's output probabilities; if the model is not well-calibrated, threshold functions set based on probability values will lose their theoretical foundation.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065217API
curl https://kongchang.com/api/v1/knowledge/claims/1065217MCP
get_claim(id=1065217)