Unverified50% confidenceFactExact time
The study finds that risk shows systematic differences across models, and these differences cannot be explained by capability level alone.
1
Sources
50%
Confidence
Long-term
Relevance
11/6/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA model's 'raw capability level' and its 'actual deployment risk' are two concepts that need to be considered separately.76% similarUnverifiedOptimizing for benchmarks carries risks, as high scores may only reflect a model's adaptation to specific question types rather than a genuine improvement in general capabilities.76% similarUnverifiedCapability assessment faces two major difficulties: inconsistent standards (conclusions measured by different organizations vary greatly) and the emergent and context-dependent nature of capabilities (a model that shows no risk today may become dangerous after fine-tuning or combination with other tools).76% similarUnverifiedThe performance differences among the general System One model, the task baselines, and the open-source fine-tuned model further point to one conclusion: improvements in model capability cannot replace risk control design at the decision layer.74% similarUnverifiedIn many casual model comparisons, the variance introduced by the harness is large enough to swamp the variance from the models themselves, though this does not mean there are no real differences between models.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1141712API
curl https://kongchang.com/api/v1/knowledge/claims/1141712MCP
get_claim(id=1141712)