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

AI可可AI生活

[人人能懂] AI在模仿、规划,还是真的在思考?

16 Sep 2025

Description

有没有想过,AI的学习之路,和我们人类有多像,又有多不像?本期节目,我们将一口气为你拆解五篇最新论文,看看AI究竟是如何学习的:它既能像婴儿一样,通过“思想实验”无师自通地理解世界;也能像一个项目总指挥,用“加杠杆”的智慧来管理庞大的训练工程。我们还会看到,AI正在从一个埋头解题的“学生”,转变为一个善用人类工具的“规划师”,而它的成长也需要一套科学的“发育课程”来避免“躺平摆烂”。最后,我们还要揭开AI“现学现卖”能力的底牌,看看它到底是天才,还是一个高明的“格式控”?00:00:43 AI的自我进化:它怎么学会了“无师自通”?00:06:14 AI训练的总指挥:为什么有时候要给下属的成果“加杠杆”?00:11:48 AI的新角色:与其当“解题者”,不如当“规划师”00:17:21 AI的“成长发育”:为什么先学会画线,才能成为大师?00:23:13 “现学现卖”的AI,真的学会了吗?本期介绍的几篇论文:[CV] World Modeling with Probabilistic Structure Integration  [Stanford University]  https://arxiv.org/abs/2509.09737  ---[LG] Understanding Outer Optimizers in Local SGD: Learning Rates, Momentum, and Acceleration  [Google Research & Google DeepMind & Princeton University]  https://arxiv.org/abs/2509.10439  ---[LG] SciML Agents: Write the Solver, Not the Solution  [UC Berkeley & LBNL]  https://arxiv.org/abs/2509.09936  ---[CV] LayerLock: Non-collapsing Representation Learning with Progressive Freezing  [Google DeepMind]  https://arxiv.org/abs/2509.10156  ---[CL] Is In-Context Learning Learning?  [Microsoft]  https://arxiv.org/abs/2509.10414  

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.