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

AI可可AI生活

[人人能懂] 开卷考试、梦境健身房与那条没走的路

08 Nov 2025

Description

今天,我们将一起探索AI那些不为人知的“内心世界”和“隐藏技能”。我们将揭示AI如何“感知”到那些它放弃了的“平行世界”,又如何区分自己是“真的不懂”还是“问题太复杂”。同时,我们还会看看它如何通过“开卷考试”和在“梦境健身房”里训练,突破我们想象的效率极限。这些最新论文,正在颠覆我们对AI效率、智能甚至“坦诚”的传统认知。00:00:34 AI加速生成:快与好的两难,如何破局?00:07:26 AI的“遗忘”与“再利用”:一份被浪费的宝藏00:12:42 AI的“内心戏”:它知道自己放弃了什么吗?00:18:00 AI的专属健身房:让它在梦里学会真本事00:23:35 AI的“我不知道”,你真的读懂了吗?本期介绍的几篇论文:[LG] Optimal Inference Schedules for Masked Diffusion Models  [Harvard & UW]  https://arxiv.org/abs/2511.04647 ---[CL] Reusing Pre-Training Data at Test Time is a Compute Multiplier  [Apple & Stanford]  https://arxiv.org/abs/2511.04234 ---[CL] Are language models aware of the road not taken? Token-level uncertainty and hidden state dynamics  [Stanford University & Goodfire & NTT Research]  https://arxiv.org/abs/2511.04527 ---[LG] Scaling Agent Learning via Experience Synthesis  [Meta Superintelligence Labs]  https://arxiv.org/abs/2511.03773 ---[LG] The Illusion of Certainty: Uncertainty quantification for LLMs fails under ambiguity  [Technical University of Munich]  https://arxiv.org/abs/2511.04418 

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.