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

AI可可AI生活

[人人能懂] 从线性内核、新词沟通到微创微调

11 Oct 2025

Description

你有没有想过,让一个复杂的AI黑箱像线性代数一样清晰可控?或者,我们根本不用动它的大脑,只要教它一个新词,就能让它和我们心有灵犀?甚至,让AI原地反复“琢磨”就能变聪明,或者只动一个“微创手术”就能完成一次完美进化?今天,我们就来聊聊这些脑洞大开的最新论文,看看科学家们如何用最聪明的“笨办法”,撬动AI的智慧极限。00:00:32 AI黑箱的新“坐标系”:当非线性遇上线性代数00:06:59 给AI造个新词,就能跟它心有灵犀?00:12:07 AI学习的“错题本”智慧00:16:47 大模型微调的新思路:为什么动大手术还不如做个“微创”?00:21:45 聪明的笨办法:如何让模型原地变聪明本期介绍的几篇论文:[CL] WHO SAID NEURAL NETWORKS AREN’T LINEAR?  [Preprint]  https://arxiv.org/abs/2306.14006 ---[CL] NEOLOGISM LEARNING FOR CONTROLLABILITY AND SELF-VERBALIZATION  [Preprint]  https://arxiv.org/abs/2401.01401 ---[CL] Agent Learning via Early Experience  [Preprint]  https://arxiv.org/abs/2309.13866 ---[CL] SLICEFINE: THE UNIVERSAL WINNING-SLICE HYPOTHESIS FOR PRETRAINED NETWORKS  [Preprint]  https://arxiv.org/abs/2307.09154 ---[CL] Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughts  [Preprint]  https://arxiv.org/abs/2305.14194 

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.