Unverified50% confidenceFactExact time
On 162 SciFact long queries, just two random terms can cover 19.9% of the eligible terms and still produce a 39.5% Hybrid target exposure rate.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAblation experiments show that, at the same precision, soft-label hierarchical retention keeps 751% more candidate words than strict power gating, expanding the candidate pool from 41 words to 349 words.71% similarUnverifiedFinite-sample numerical experiments show that under both settings, the sample risk premium accounts for about 74% to 97% of the total regret.67% similarUnverifiedTheoretically reliable clustered standard error estimation typically requires at least 30–50 clusters, whereas this study has only 19 independent time units, making the sample size relatively small.67% similarUnverifiedAmong all 13 models, only 37% of samples with label changes shifted toward the misleading image's meaning, while 63% went in the opposite direction, both below the 50% random baseline, indicating that images merely randomly disrupted model judgments rather than guiding them.67% similarUnverifiedAccording to official Anthropic data, when the "self-evaluation" system was tested on Claude Fable 5, it identified 78% of potentially harmful conversations and 94% of high-priority cases flagged by a stronger classifier, with a false positive rate of only 1%.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1133895API
curl https://kongchang.com/api/v1/knowledge/claims/1133895MCP
get_claim(id=1133895)