Unverified50% confidenceFactExact time
Research shows that a model's accuracy on self-criticism tasks is significantly lower than in scenarios where criticism is performed by another model.
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
Codex插件接入Claude Code：双Agent协作审查代码的新范式
bilibili锋芒AI7/12/2026
Related Claims
UnverifiedAnthropic's research shows that even mild alignment training can cause a 2–8 percentage point decline in model performance on certain reasoning-intensive tasks.76% similarUnverified在事实性评估任务上模型评判者准确率仍显著低于人类专家，但在风格、连贯性等主观维度已接近人类评审员间一致性水平76% similarUnverified不同模型在基准测试上的性能差异往往远小于同一模型在不同任务类型、不同输入格式下的行为差异70% similarUnverified研究表明在超过8-10轮对话后，模型对初始系统提示中规则的遵循率可下降20%-40%69% similarUnverified斯坦福大学研究发现，在推理链关键步骤注入错误信息对最终答案的影响远小于预期69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/606979API
curl https://kongchang.com/api/v1/knowledge/claims/606979MCP
get_claim(id=606979)