![[人人能懂AI前沿] 从递归自省、二维视角到神经动力学](https://image.xyzcdn.net/FuDP4HpAp8ezgVZMmEel3mblKCmJ.jpg@middle)
主播
节目简介
来源：小宇宙
你有没有想过，AI和我们人类一样，也需要一套“成长方法论”？本期节目，我们就来聊聊几篇最新论文揭示的AI高手修炼秘籍：它们不仅要纠结是先上“通识课”还是先搞“专才特训”，甚至连最简单的复制粘贴都做不好，需要通过“认知升维”来解决。我们还会发现，一本好的“工作手册”可能比模型本身更重要，而AI“脑补”世界的方式，竟然和我们的大脑惊人地相似。最后，我们会看到AI如何学会“把力气用在刀刃上”的做事智慧。
00:00:36 AI高手是怎样炼成的，通才教育还是专才特训？
00:06:20 为什么顶尖模型连复制粘贴都做不好？
00:11:11 比模型更重要的，可能是它的“工作手册”
00:16:31 你的大脑，如何看穿了AI的秘密？
00:24:07 做事高手的方法论，如何把力气用在刀刃上？
本期介绍的几篇论文：
[LG] Understanding Reasoning from Pretraining to Post-Training
[New York University]
https://arxiv.org/abs/2607.16097
---
[CL] Frontier Language Models Struggle to Copy: Text Can Be Better Viewed in 2D
[Tsinghua University]
https://arxiv.org/abs/2607.16072
---
[LG] Recursive Harness Self-Improvement
[Sakana AI & UC Berkeley]
https://arxiv.org/abs/2607.15524
---
[AI] Toward a mechanistic understanding of inference in visual cortex and diffusion models
[UC Berkeley]
https://arxiv.org/abs/2607.15693
---
[CL] Process Reward Informed Tree Rollout for Effective Multi-Turn RL
[UC San Diego & Amazon]
https://arxiv.org/abs/2607.15610
00:00:36 AI高手是怎样炼成的，通才教育还是专才特训？
00:06:20 为什么顶尖模型连复制粘贴都做不好？
00:11:11 比模型更重要的，可能是它的“工作手册”
00:16:31 你的大脑，如何看穿了AI的秘密？
00:24:07 做事高手的方法论，如何把力气用在刀刃上？
本期介绍的几篇论文：
[LG] Understanding Reasoning from Pretraining to Post-Training
[New York University]
https://arxiv.org/abs/2607.16097
---
[CL] Frontier Language Models Struggle to Copy: Text Can Be Better Viewed in 2D
[Tsinghua University]
https://arxiv.org/abs/2607.16072
---
[LG] Recursive Harness Self-Improvement
[Sakana AI & UC Berkeley]
https://arxiv.org/abs/2607.15524
---
[AI] Toward a mechanistic understanding of inference in visual cortex and diffusion models
[UC Berkeley]
https://arxiv.org/abs/2607.15693
---
[CL] Process Reward Informed Tree Rollout for Effective Multi-Turn RL
[UC San Diego & Amazon]
https://arxiv.org/abs/2607.15610
外观
