Unverified50% confidenceFactExact time
Correlated bias may exist between the judge and the model, i.e., the judge and the model share the same kind of blind spots or preferences, so the judge's high scores cannot independently prove the model's safety.
1
Sources
50%
Confidence
Long-term
Relevance
11/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe study finds that risk shows systematic differences across models, and these differences cannot be explained by capability level alone.77% similarUnverifiedOfficial guidance recommends that the judge model should differ from the baseline model, to avoid a model scoring its own outputs and thereby favoring outputs that are closer to its own style.75% similarUnverifiedIf a model mistakes preferences for hard constraints, it will over-filter and the recommendation coverage rate will drop; conversely, if it treats hard constraints as preferences, it will give places that are actually unusable.74% similarUnverifiedModel behavior depends on context, and a constraint that blocks the model in one workflow may be exactly the guardrail needed in another scenario, so there is no unified "prompt toxicity" theory for a command to detect.73% similarUnverifiedThe study points out a trade-off between safety and verisimilitude, where an overly broad definition of harmlessness makes models behave clumsily in adversarial scenarios.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1173869API
curl https://kongchang.com/api/v1/knowledge/claims/1173869MCP
get_claim(id=1173869)