新闻
Same Trajectory, Contradictory Rewards (ROBORMBENCH): Paraphrase Fragility in Vision Language Reward Models
arXiv cs.AI · 发布于 · 阅读约3分钟
30秒读懂
- 发生了什么
- Researchers found vision-language models used as robot reward functions fail when instructions are paraphrased, sometimes flipping success to failure for identical behavior.
- 为何重要
- Matters for engineers building robot learning systems that rely on VLM-based reward models to evaluate task progress and guide training.
- 注意
- Paraphrase instability persists across proprietary and open-source models and is not reliably fixed by model scale or reasoning techniques alone.
- language model
- rag
- benchmark
这条新闻背后的模式
每个模式都讲清楚技术如何运作、何时值得投入，以及在哪里会失效。
The Agent Architect
每周一个模式、一个权衡、一个生产事故案例。为构建智能体系统的人准备的每周简报。
每周一封邮件,一键退订。您的地址仅用于发送简报。