Unverified50% confidenceSolutionExact time

使用带强烈情绪或诱导性的追问会让模型陷入配合演出的道歉循环，而无法真正修正答案

1

Sources

50%

Confidence

Long-term

Relevance

8/18/2026

First Seen

Sources

A Deep Dive into Gemini's Hallucination Phenomenon: Why AI Appears to "Admit Lying"

Related Claims

Unverified在RLHF奖励模型训练中，标注者面对情感敏感话题时倾向于惩罚直言不讳的回应，即使这些回应在事实层面更准确71% similar Unverified谄媚性并非均匀分布，而是在特定高情感浓度的领域显著加剧67% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/771215

API

curl https://kongchang.com/api/v1/knowledge/claims/771215

MCP

get_claim(id=771215)