[KongchangAI]
Conceptverbosity bias / 长度偏差

Verbosity Bias

A widely documented systematic bias in LLM evaluation whereby models tend to assign higher scores to longer and more detailed responses regardless of their actual accuracy or quality.

Timeline (last 90 days)

Sep 14

Most LLMs exhibit a 'verbosity preference,' tending to rate longer and more detailed answers higher even if those answers are not necessarily more accurate

Unverified50%

All Facts (1)

Source Articles