[人人能懂] 从二元奖惩、动态审计到思维积木 episode artwork

EPISODE · Oct 21, 2025 · 26 MIN

[人人能懂] 从二元奖惩、动态审计到思维积木

from AI可可AI生活

本期我们要聊一个核心问题:我们总觉得AI是个神秘的黑箱,但最新的研究正在像做“脑部扫描”一样,试图撬开它。我们将看到,一个“满分或零分”的简单规则,就能教会AI诚实;又如何派出一个“AI侦探”,揪出潜伏的恶意模型。接着,我们会深入AI的“思考过程”,看看聪明的“大脑”和聪明的“搜索引擎”哪个更重要,以及如何让AI通过“犯错”来演化出正确答案,甚至把它的复杂推理拆解成一个个可以遥控的“思想积木”。准备好了吗?让我们一起深入AI的内心世界。00:00:41 AI的“不说谎”训练:满分或零分00:05:29 AI界的“无间道”:如何揪出披着羊皮的狼?00:10:39 聪明的大脑,和聪明的搜索引擎,哪个更重要?00:16:14 犯错没关系,只要你“改对”的概率比“改错”大一点点00:21:22 拆解AI大脑:它思考时在想什么?本期介绍的几篇论文:[CL] Train for Truth,Keep the Skills:Binary Retrieval-Augmented Reward Mitigates Hallucinations  [University of Washington & Allen Institute for AI (Ai2)]  https://arxiv.org/abs/2510.17733 ---[LG] Detecting Adversarial Fine-tuning with Auditing Agents  [Anthropic]  https://arxiv.org/abs/2510.16255 ---[LG] Prior Makes It Possible:From Sublinear Graph Algorithms to LLM Test-Time Methods  [Toyota Technological Institute at Chicago & Columbia University & Google Research]  https://arxiv.org/abs/2510.16609 ---[CL] Deep Self-Evolving Reasoning  [Microsoft Research Asia & Peking University]  https://arxiv.org/abs/2510.17498 ---[LG] Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models  [Columbia University & University of California Los Angeles & Harvey Mudd College]  https://arxiv.org/abs/2510.15987 在小宇宙查看该单集文稿

Episode metadata supplied by the publisher feed · Published Oct 21, 2025

Embed this episode

Ready to play

[人人能懂] 从二元奖惩、动态审计到思维积木

0:00 26:58

No transcript for this episode yet

We transcribe on demand. Request one and we'll notify you when it's ready — usually under 10 minutes.

No similar episodes found.

Frequently Asked Questions

How long is this episode of AI可可AI生活?

This episode is 26 minutes long.

When was this AI可可AI生活 episode published?

This episode was published on October 21, 2025.

Can I download this AI可可AI生活 episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.
URL copied to clipboard!