Unverified50% confidenceFactExact time
The English-trained class-mean readout outperformed both zero-shot null hypotheses in 8 of 12 discovery tests, and a stronger source-selection penalty improved transfer via logistic regression and LEACE directions.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe paper proves that, under appropriate assumptions, relative to the target amplitude, the amplitude of the learned solution at high frequencies consistently tends to zero throughout training.70% similarUnverifiedIn a test covering 8 local models and more than 5000 calls, the wording style of the prompt (plain/normal/florid) had almost no effect on accuracy on non-math tasks, with differences of no more than 4 percentage points for models of 8B and above.69% similarUnverifiedWhen sampling eight candidates without constraints from the model, 10.2% of inputs cannot produce any valid output under limited training data, and this proportion drops to 0.4% when using the full training data68% similarUnverifiedThe original confirmatory logistic regression family rejected both planned null hypotheses in 4 of 12 tests, with the English direction exceeding the majority-class baseline by only 3.0-7.3 percentage points.68% similarUnverifiedBefore testing, the semantic threshold was fixed at 0.81, a value calibrated on the Banking77 training set to guarantee that at least 98% of semantic hits point to the same intent; but on the actual test set the intent agreement rate of semantic hits was 96.7% to 96.8%, with each model having 30 to 31 hits on the wrong intent.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1153329API
curl https://kongchang.com/api/v1/knowledge/claims/1153329MCP
get_claim(id=1153329)