Unverified50% confidenceBenchmarkExact time
On a dataset containing 45 drivers, TEMPEST achieves a Rank-1 accuracy of 91.71% under strict temporal evaluation.
1
Sources
50%
Confidence
Long-term
Relevance
10/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe metrics of over 90% accuracy and completion within one day were achieved in specific scenarios (such as visual reasoning at traffic intersections), and actual results vary depending on data quality and task complexity.65% similarUnverifiedThe best F1 scores for sarcasm detection tasks on standard datasets (such as iSarcasm and the SemEval series) have long hovered in the 70%–80% range.62% similarUnverifiedIn the Phase 1 post-hoc comparison, SAGE's evaluation results were compared against 105 human reviews from the original competition, yielding an ordinal consistency of kappa = 0.29, which falls in the "fair" range.62% similarUnverifiedAn audit of thousands of agent execution trajectories in DeepSWE-1.1 found that more than 80% of the trajectories contained reasoning about an "imagined grader."61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/983176API
curl https://kongchang.com/api/v1/knowledge/claims/983176MCP
get_claim(id=983176)