你有没有想过,我们衡量AI的标准可能从一开始就有点偏?今天,我们将一起颠覆几个常识:我们会发现,模型的潜力不在于考高分,而在于它的“想象力”有多丰富;训练巨型模型的省钱秘笈,可能就藏在一个简单的根号里;而要让AI生成完美的视频,最好的方法竟是让它组建一个内部“评审团”给自己挑错。更神奇的是,想让AI真正懂你,关键或许不是问“二选一”,而是“三选一”。准备好了吗?让我们一起探索这些最新论文中,那些反直觉又充满智慧的AI新思路。00:00:41 AI训练的内功心法:为什么“好学生”不一定是“优等生”?00:07:18 炼大模型省钱秘笈:一个根号引发的蝴蝶效应00:12:05 让AI自己挑毛病,视频才能越做越好00:17:25 想真正懂我?别问二选一,试试三选一00:21:57 给AI装个“健康码”:识别未知攻击的新思路本期介绍的几篇论文:[LG] The Coverage Principle: How Pre-training Enables Post-Training[Microsoft Research & MIT & UIUC]https://arxiv.org/abs/2510.15020---[LG] Robust Layerwise Scaling Rules by Proper Weight Decay Tuning[MIT & UCLA]https://arxiv.org/abs/2510.15262---[CV] VISTA: A Test-Time Self-Improving Video Generation Agent[Google]https://arxiv.org/abs/2510.15831---[LG] Learning Correlated Reward Models: Statistical Barriers and Opportunities[MIT EECS]https://arxiv.org/abs/2510.15839---[CV] Learning to Detect Unknown Jailbreak Attacks in Large Vision-Language Models[Renmin University of China & Alibaba Group]https://arxiv.org/abs/2510.15430
No persons identified in this episode.
This episode hasn't been transcribed yet
Help us prioritize this episode for transcription by upvoting it.
Popular episodes get transcribed faster
Other recent transcribed episodes
Transcribed and ready to explore now
SpaceX Said to Pursue 2026 IPO
10 Dec 2025
Bloomberg Tech
Don’t Call It a Comeback
10 Dec 2025
Motley Fool Money
Japan Claims AGI, Pentagon Adopts Gemini, and MIT Designs New Medicines
10 Dec 2025
The Daily AI Show
Eric Larsen on the emergence and potential of AI in healthcare
10 Dec 2025
McKinsey on Healthcare
What it will take for AI to scale (energy, compute, talent)
10 Dec 2025
Azeem Azhar's Exponential View
Reducing Burnout and Boosting Revenue in ASCs
10 Dec 2025
Becker’s Healthcare -- Spine and Orthopedic Podcast