Towards Data Science· Abdullahi Dattijo·· 3 小时前AI 评分56
AI智能体如何响应错误奖励标签：GPT-6 Sol、Claude Haiku与Qwen3的对比实验
I Told an AI Agent That Its Wrong Answers Were Right. One Model Believed Me.
AI 导读
作者设计了一个隐藏规则的路由选择任务，测试GPT-6 Sol、Claude Haiku 5.5和Qwen3 4B Instruct在真实、翻转、随机等不同奖励标签下的表现。
来源：Towards Data Science · towardsdatascience.com