GPT's Words Are Too Safe to Read

“Like humans, LLMs will often give junk feedback. In my experience, OpenAI models will always tell me to moderate my claims or add caveats and hedges until I’m not saying anything at all.”

—— Sean Goedecke, Blog about things you don’t understand yet

GPT 倾向于在一个可能有争议性的 claim 上加上很多保守表述，最后让一个 claim 完全无法识别或者被淹没。

类似于正确的废话（“it depends”）。一个安全而永远不会错的假设，无法提供有效预测，也就不包含知识。