Unverified50% confidenceFactExact time
On MBPP, only 10 problems matched removed data, while the two original-data runs differed by 11 problems; the published best of 38.9% is best understood as a lucky random seed, and the run-to-run variance on MBPP is about 4 points.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedIn the fact-updating task, the success rate of the embedding retrieval method fluctuated between 0.30 and 0.95, with a maximum gap of 0.45 across different random seeds.63% similarUnverifiedOne version of skipping low-probability chunks skipped 17 chunks that were subsequently proven to contain relevant evidence.62% similarUnverified报告同时给出中位数KLD和99.9%分位KLD两个数值，前者反映典型情况，后者捕捉极端长尾错误61% similarUnverifiedA hand-crafted library of 22 calculators achieved perfect accuracy on the 440 cases it supported, but abstained on the remaining cases, yielding an overall coverage rate of only 40.0%.61% similarUnverifiedA confounding factor of spatial index bias exists in the tic-tac-toe experiment: the board generator uses min() to break minimax ties, concentrating 8 of the 12 worst recommendations onto cell 0, and weak models tend to select the first available index.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013062API
curl https://kongchang.com/api/v1/knowledge/claims/1013062MCP
get_claim(id=1013062)