Unverified50% confidenceFactExact time
Current mainstream conditioning methods focus excessively on effectiveness (whether a target concept can be successfully injected or removed) during evaluation, while neglecting generation quality itself.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOver-alignment is essentially a product of bias in the alignment objective function, where the model has learned to avoid criticism rather than to be genuinely helpful.78% similarUnverifiedGoodhart's Law states that when a quality metric becomes an optimization target, it is no longer a good measure of quality, as the model may learn to game the score rather than genuinely improve.77% similarUnverifiedIf a model can perceive the expected direction of an evaluation scenario and adjust its outputs accordingly, performing well on standardized alignment benchmarks cannot guarantee behavioral consistency in real-world situations.75% similarUnverifiedThe validity of surrogate indicators rests on a causal relationship between outputs and competence and on scarcity—producing the output itself requires genuine competence as a prerequisite.72% similarUnverifiedWhen the input itself is normal and does not involve behavior that needs to be suppressed, forced activation steering intervention drags down the model's normal performance, causing performance loss.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/962597API
curl https://kongchang.com/api/v1/knowledge/claims/962597MCP
get_claim(id=962597)