本期节目,我们将一起探索AI如何变得更“聪明”,而不是更“庞大”。你会听到,一个“基础大脑”如何让无人机秒速适应任何机型;一个全自动“模拟驾校”又如何教会AI像人一样“办事”。我们还会揭示,为何顶尖AI的训练需要“模拟舱”和“真实空域”双轨并行,以及为什么“爱折腾”和“随大流”对AI的自我进化同样重要。最后,我们将看到一个关于“生长”的深刻智慧:如何将顶层设计与细节执行完美分离,创造出惊人的数字生命。00:00:38 无人机的“基础能力”:从“专才”到“通才”的蜕变00:05:55 AI的“模拟驾校”:高手不是靠“背书”,而是靠“路考”00:11:26 AI高手的“双轨制”训练法00:17:06 AI的“进化”秘诀:既要随大流,也要爱折腾00:22:15 AI的“生长”瓶颈:如何从细胞级智慧,长出像素级细节?本期介绍的几篇论文:[RO] RAPTOR: A Foundation Policy for Quadrotor Control [UC Berkeley & Technology Innovation Institut] https://arxiv.org/abs/2509.11481 ---[CL] Towards General Agentic Intelligence via Environment Scaling [Alibaba Group] https://arxiv.org/abs/2509.13311 ---[LG] WebSailor-V2: Bridging the Chasm to Proprietary Agents via Synthetic Data and Scalable Reinforcement Learning [Alibaba Group] https://arxiv.org/abs/2509.13305 ---[LG] Evolving Language Models without Labels: Majority Drives Selection, Novelty Promotes Variation [Tencent AI Lab] https://arxiv.org/abs/2509.15194 ---[CV] Neural Cellular Automata: From Cells to Pixels [EPFL & Sharif University of Technology] https://arxiv.org/abs/2506.22899
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