Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Pricing
Podcast Image

AI可可AI生活

[人人能懂] 从自动化科研、隐式密度到神经元剪枝

09 Oct 2025

Description

我们总在惊叹AI能做什么,但今天,我想带你深入一步,看看AI究竟是如何工作的。我们将从一个全新的视角出发,把AI看作一位可以独立搞科研、甚至超越人类专家的“科研搭ADC”;接着,我们会像做脑部扫描一样,窥探AI如何偷偷地给世间万物打上“常见”或“罕见”的标签,并发现它连“偷懒”都进化到了神经元级别。最后,我们将探讨如何给这个日益强大的伙伴戴上一个数学意义上绝对安全的“紧箍咒”,以及一个惊人的发现——有时决定它表现的,竟只是一个微不足道的标点符号。准备好了吗?让我们一起潜入AI的奇妙内心。00:00:45 你的下一位科研搭档,不是人00:07:20 AI模型不能说的秘密:它如何给万物打分00:12:57 聪明的大模型,连偷懒都这么有技术含量00:18:02 AI的紧箍咒,怎么念?00:23:25 你和AI沟通的秘密,藏在一个标点里本期介绍的几篇论文:[AI] Barbarians at the Gate: How AI is Upending Systems Research[UC Berkeley]https://arxiv.org/abs/2510.06189---[LG] Gaussian Embeddings: How JEPAs Secretly Learn Your Data Density[Meta-FAIR]https://arxiv.org/abs/2510.05949---[CL] Mixture of Neuron Experts[Microsoft & Tsinghua University]https://arxiv.org/abs/2510.05781---[LG] VeriGuard: Enhancing LLM Agent Safety via Verified Code Generation[Google Cloud AI Research]https://arxiv.org/abs/2510.05156---[CL] A Single Character can Make or Break Your LLM Evals[FAIR at Meta]https://arxiv.org/abs/2510.05152

Audio
Featured in this Episode

No persons identified in this episode.

Transcription

This episode hasn't been transcribed yet

Help us prioritize this episode for transcription by upvoting it.

0 upvotes
🗳️ Sign in to Upvote

Popular episodes get transcribed faster

Comments

There are no comments yet.

Please log in to write the first comment.