EPISODE · Feb 20, 2026 · 27 MIN
[人人能懂AI前沿] AI的“悟性”:从多视角脑补、系统化草稿到渐进式遗忘
from AI可可AI生活
你有没有想过,AI要怎样才能像我们一样,“脑补”出世界的立体样貌,而不是死记硬背照片?当AI进行创作时,它脑中的“草稿”怎样才算恰到好处?本期节目,我们将一起探索几篇最新论文,看看AI如何通过理解“关系”而非事实获得视觉,如何通过引入“陪练老师”和“参考答案”实现自我进化,甚至学会了将记忆内化为“肌肉记忆”的“断舍离”神功,以及像搭乐高一样构建自己的能力。00:00:34 你的大脑如何“脑补”出整个世界?00:07:02 AI绘画的“草稿”,怎么打才算恰到好处?00:12:32 为什么最聪明的AI,要先学会“断舍离”?00:17:38 乐高式AI,未来模型的正确拼法00:23:06 AI进化的秘密,优等生的“参考答案”本期介绍的几篇论文:[CV] Human-level 3D shape perception emerges from multi-view learning [UC Berkeley] https://arxiv.org/abs/2602.17650 ---[LG] Unified Latents (UL): How to train your latents [Google DeepMind Amsterdam] https://arxiv.org/abs/2602.17270 ---[LG] Training Large Reasoning Models Efficiently via Progressive Thought Encoding [Microsoft Research & University of Rochester] https://arxiv.org/abs/2602.16839 ---[LG] A Theoretical Framework for Modular Learning of Robust Generative Models [Google Research] https://arxiv.org/abs/2602.17554 ---[CL] References Improve LLM Alignment in Non-Verifiable Domains [Yale University & Meta] https://arxiv.org/abs/2602.16802 在小宇宙查看该单集文稿
Embed this episode
Ready to play
[人人能懂AI前沿] AI的“悟性”:从多视角脑补、系统化草稿到渐进式遗忘
No transcript for this episode yet
Similar Episodes
No similar episodes found.