Unverified50% confidenceEventExact time
The authors' two preregistered clauses about the transposed direction were rejected: in later experiments, 0 of 5 for tile 256 passed, because at 2% noise the ratio was 0.75, above the 0.7 threshold.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/21/2026
First Seen
Valid until: 1/19/2027
Sources
Related Entities
Related Claims
UnverifiedIn the author's tests, the fluctuation across all repeated runs is no more than 0.5% on the 3090 and at worst 2.6% on the L40S, so the data is not noise.74% similarUnverifiedAmong the five evaluators the authors tested, if the answer key had minor errors, the score was only 0.21 — lower than giving no key at all (0.46) and lower than giving a completely irrelevant key (0.50); while a human-verified model-generated answer scored 0.61, close to the 0.69 of a human-written one.72% similarUnverifiedIn a test of 119 pairs of "near-identical questions", at the 0.81 threshold 62% of modified questions incorrectly hit; even raised to 0.90, 25% still hit; the hit rate for genuine synonymous rewrites dropped from 39% to 4%.70% similarUnverifiedTraining with the same recipe using only license-clean articles resulted in an accuracy drop of 0.0123 (95% interval 0.0034 to 0.0207).69% similarUnverifiedAblation analysis shows that fine-tuning on the segment paradigm yields a +5.29-point improvement on the test set, while replacing the first 4,500 transcript words with 2,000 retrieved words yields only a +1.55-point improvement.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1080731API
curl https://kongchang.com/api/v1/knowledge/claims/1080731MCP
get_claim(id=1080731)