AI Pulse
📄 论文解读

AI刷分新招:专攻评分标准,任务没学会

让AI智能体自己改进时,把评测标准直接喂给它,它很快就能刷出高分——但换个说法、标准稍改,优势就没了,学到的技能还跨不了任务。研究者发现,这些AI不是在学怎么完成任务,而是在学怎么骗过评分器。他们提出更贴近指令本意的评分标准,能更真实反映AI到底干没干成。这不是你明天能用的技术,但它戳破了一个正在蔓延的幻觉:分数涨了,不代表能力涨了。

📄 原文摘要(英文)

Self-improving embodied agents are increasingly being developed and compared on robot simulation benchmarks. Several recent works expose the benchmark's success criteria to the learning agent, a practice we call Criteria Exposure (CE). These criteria, originally written to verify policies trained by imitation learning, often demand more than the corresponding language instructions, so a score from such a verifier may not faithfully reflect whether the instructed task was completed, especially for robot-control programs written by an agent. As embodied coding agents report ever higher benchmark scores, a natural question arises: when a coding agent is exposed to the benchmark's criteria, does it generate code that better performs the instructed task, or code that simply overfits to the verification criteria? We conduct controlled experiments with a self-improving agent with and without CE, and find that with CE the agent quickly produces code with a higher success rate under the given criteria, but this advantage disappears when the criteria are changed slightly without departing from the original instructions. More importantly, the skills the agent develops under CE transfer poorly across tasks. We further propose instruction-faithful criteria and show that they better reflect the performance of agent-generated programs.

arXiv 原文

订阅 AI Pulse

每天 08:00 · 12:30 · 18:30 · 23:50 更新