Unverified50% confidenceFactExact time
Explicitly providing temporal information in the prompt resolves only about half of the failure cases.
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedApproximately two-thirds of failed trajectories are 'near-misses.'69% similarUnverifiedThe failure-replay mechanism is only effective when measured in "steps" rather than "games"; the first version replayed a quarter of new games, but the replayed data accounted for less than 1% of the total training volume.68% similarUnverifiedDue to sampling variance, each test message should be run twice; a rule that holds with only 50% probability is effectively no rule at all.64% similarUnverifiedThe ReAct paper reports that on questions requiring lookups, it reduced fabricated facts produced by pure reasoning by about half.63% similarUnverifiedOn a small test set of only one hundred scenarios, scores can swing by several percentage points between different runs, and judgment should be made on the full dataset.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1139003API
curl https://kongchang.com/api/v1/knowledge/claims/1139003MCP
get_claim(id=1139003)