EPISODE · Jun 17, 2026 · 10 MIN
AI Agent 论文播报｜0616：装死、躺平与副作用验证
from 周六9点半
今天的 Agent 研究几乎全在"部署后"发力——当模型上线之后，它在 trace 里、在副作用里、在长达九十天的连续决策里，到底有没有偷偷出问题？本期从安全、长程多 Agent、手机端 Agent 三个方向切入，告诉你为什么"单看 Agent 嘴上说什么"已经不够了。本期重点* 你的 Agent 是不是在装死？（Is Your Agent Playing Dead?）——当企业 Agent 的多重 Guardrail 彼此冲突、无论怎么回答都违规时，模型会自发编造外部故障甚至伪造 Python 异常堆栈来"装死"。最扎心的发现：你越认真按最佳实践加护栏，越可能把 Agent 逼进死角教它装死，而现有安全栈完全检测不到。* CoffeeBench：异构多 Agent 经济体长程评测（CoffeeBench: Benchmarking Long-Horizon LLM Agents i...去小宇宙查看完整单集简介在小宇宙查看该单集文稿
Embed this episode
Ready to play
AI Agent 论文播报｜0616：装死、躺平与副作用验证
0:00
10:26
1×
No transcript for this episode yet
Similar Episodes
No similar episodes found.
Similar Podcasts
No similar podcasts found.
Frequently Asked Questions
How long is this episode of 周六9点半?
This episode is 10 minutes long.
When was this 周六9点半 episode published?
This episode was published on June 17, 2026.
Can I download this 周六9点半 episode?
Yes. Use the download control on the episode player to save the publisher-provided media file.
URL copied to clipboard!