Conceptover-refusal / 过度拒绝

Over-Refusal

Refers to the behavior of large language models refusing harmless requests that should normally be answered. Together with safety rate, it forms a tension metric that jointly characterizes the safety-usefulness frontier.

Timeline (last 90 days)

Oct 15

In the controlled sweep of the general-data proportion in safety SFT, the over-refusal curve is non-monotonic, with the inflection point falling in the 40%–50% general-data proportion range.

Unverified50%

All Facts (1)

Unverified

In the controlled sweep of the general-data proportion in safety SFT, the over-refusal curve is non-monotonic, with the inflection point falling in the 40%–50% general-data proportion range.

50%

Source Articles

Manacá-1B-Instruct-2: Reproducible Safety Alignment for Brazilian Portuguese LLMs and a DPO Negative ResultOct 15