Unverified50% confidenceFactExact time
The weighted upper confidence limit (UCL) of expected harm was 4.60%, but only 2 of 11 preset margins passed the support/risk rules.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/11/2026
First Seen
Valid until: 2/9/2027
Sources
Related Entities
Related Claims
UnverifiedThe identification and refusal of compositional harm are decoupled: nine standard direction-finding methods detect compositional harm with an AUC of up to 0.99, but ablating these directions either does not affect refusal behavior or destroys model coherence.64% similarUnverifiedIn a head-to-head comparison with four external PT-BR models, the lower bound of the Wilson confidence interval for the safety rate of Manacá-1B-Instruct-2 is 94.1%, higher than the upper bounds of all four baseline models.61% similarUnverifiedFinite-sample numerical experiments show that under both settings, the sample risk premium accounts for about 74% to 97% of the total regret.58% similarUnverifiedThe separately tested version with the auxiliary danger head reached 99.7% recall but still missed a rare straight-line configuration, resulting in a confident collision.57% similarUnverifiedSimulated control evaluations (SCEs) predicted safety with a root mean square error of 3.6 percentage points (at a 5% audit budget and under five monitor conditions), compared with 10.1 percentage points for BashArena.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1158205API
curl https://kongchang.com/api/v1/knowledge/claims/1158205MCP
get_claim(id=1158205)