Unverified50% confidenceFactExact time
Overly tightened guardrails lead to the phenomenon of over-refusal; research indicates that models trained with strict RLHF alignment may have refusal rates 3–5 times higher than actually necessary.
1
Sources
50%
Confidence
Long-term
Relevance
9/4/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedExcessive safety fine-tuning can cause models to over-refuse, undermining their practical utility.77% similarUnverifiedDrift detection involves a trade-off: setting the threshold too sensitively leads to frequent false alarms and unnecessary retraining, while setting it too loosely may allow model performance to continuously degrade without being detected.76% similarUnverifiedOver-alignment or over-refusal is a common engineering challenge faced by current mainstream large language models, where models are trained to be overly conservative, resulting in a high false-positive rate for reasonable requests.76% similarUnverifiedThe alignment tax refers to the phenomenon where, when a model is overtrained to refuse harmful requests, its refusal boundary may generalize to normal harmless scenarios, resulting in excessive caution.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/854288API
curl https://kongchang.com/api/v1/knowledge/claims/854288MCP
get_claim(id=854288)