![[人人能懂AI前沿] AI的成长三部曲：金牌教练、乐高大师与风光摄影师](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如何像我们一样，在反复试错后找到“刚刚好”的平衡点？这一期，我们就从几篇最新的AI论文出发，聊聊AI的“智慧进化”：看它如何学会给自己配备一个“后悔调节器”来动态调整策略，如何通过带“复盘笔记”的刻意练习，从沟通“小白”进化成“流程大师”，以及如何像拼乐高一样，用聪明的设计给自己“瘦身”，最终实现速度与质量的完美飞跃。
00:00:31 做对选择，你需要一个“后悔调节器”
00:05:28 AI 进化论，如何让一个聪明的“员工”，听懂“人话”？
00:11:15 面对海量选择，我们如何做出“刚刚好”的聪明决策？
00:19:09 AI作画提速的秘密，多看一步，不止平均
00:24:26 神经网络的大瘦身，为什么聪明的设计胜过蛮力计算？
本期介绍的几篇论文：
[LG] Efficient Online Conformal Selection with Limited Feedback
[Google Research & Duke University]
https://arxiv.org/abs/2605.14953
---
[LG] Prompting Policies for Multi-step Reasoning and Tool-Use in Black-box LLMs with Iterative Distillation of Experience
[Google Research]
https://arxiv.org/abs/2605.14443
---
[LG] Stochastic Matching via Local Sparsification
[Google Research]
https://arxiv.org/abs/2605.14195
---
[LG] Covariance-aware sampling for Diffusion Models
[Google]
https://arxiv.org/abs/2605.13910
---
[LG] Compositional Sparsity as an Inductive Bias for Neural Architecture Design
[University College London]
https://arxiv.org/abs/2605.14764
00:00:31 做对选择，你需要一个“后悔调节器”
00:05:28 AI 进化论，如何让一个聪明的“员工”，听懂“人话”？
00:11:15 面对海量选择，我们如何做出“刚刚好”的聪明决策？
00:19:09 AI作画提速的秘密，多看一步，不止平均
00:24:26 神经网络的大瘦身，为什么聪明的设计胜过蛮力计算？
本期介绍的几篇论文：
[LG] Efficient Online Conformal Selection with Limited Feedback
[Google Research & Duke University]
https://arxiv.org/abs/2605.14953
---
[LG] Prompting Policies for Multi-step Reasoning and Tool-Use in Black-box LLMs with Iterative Distillation of Experience
[Google Research]
https://arxiv.org/abs/2605.14443
---
[LG] Stochastic Matching via Local Sparsification
[Google Research]
https://arxiv.org/abs/2605.14195
---
[LG] Covariance-aware sampling for Diffusion Models
[Google]
https://arxiv.org/abs/2605.13910
---
[LG] Compositional Sparsity as an Inductive Bias for Neural Architecture Design
[University College London]
https://arxiv.org/abs/2605.14764
外观
