![[人人能懂AI前沿] 给AI装上“第六感”：从感知、记忆到协作的全面升级](https://image.xyzcdn.net/FuDP4HpAp8ezgVZMmEel3mblKCmJ.jpg@middle)
主播
节目简介
来源：小宇宙
你有没有想过，未来的AI不再只是一个“大力士”，而是一个善用巧劲的“策略家”？本期节目，我们将一起探寻几篇最新论文，看看AI如何学会像我们一样“抓重点”来读懂万字长文，又如何通过一个“共享白板”实现高效团队协作。我们还会揭秘，AI怎样无中生有地获得“第六感”，如何靠着一本“记忆草稿本”创作出不会失忆的长视频，以及它又是怎样得到一支“魔法棒”，能对你的文章进行外科手术般的精准修改。准备好了吗？让我们一起看看AI是如何变聪明的！
00:00:39 你的手机，是怎么“偷懒”变聪明的？
00:05:03 人多不一定力量大，但“会开会”的团队可以
00:11:33 给机器装上“第六感”，一种更聪明的学习方式
00:18:02 AI做长视频，怎样才能不“失忆”？
00:23:18 给AI一支“局部修改”的魔法棒
本期介绍的几篇论文：
[LG] MiniMax Sparse Attention: MiniMax is a surprisingly effective reward for instruction following
[MiniMax]
https://arxiv.org/abs/2606.13392
---
[LG] Decentralized Multi-Agent Systems with Shared Context: Decentralized Multi-Agent Systems with Shared Context
[Stanford University]
https://arxiv.org/abs/2606.10662
---
[RO] FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning: FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning
[CMU]
https://arxiv.org/abs/2606.12406
---
[CV] MilliVid: Hierarchical Latents for Long-Range Consistency in Video Generation: MilliVid: Hierarchical Latents for Long-Range Consistency in Video Generation
[MIT & Toyota Research Institute]
https://arxiv.org/abs/2606.09056
---
[CL] TimpaTeks: Automatic In-place Text Sequence Modification via Diffusion Language Model Steering: TimpaTeks: Automatic In-place Text Sequence Modification via Diffusion Language Model Steering
[MBZUAI]
https://arxiv.org/abs/2606.08408
00:00:39 你的手机，是怎么“偷懒”变聪明的？
00:05:03 人多不一定力量大，但“会开会”的团队可以
00:11:33 给机器装上“第六感”，一种更聪明的学习方式
00:18:02 AI做长视频，怎样才能不“失忆”？
00:23:18 给AI一支“局部修改”的魔法棒
本期介绍的几篇论文：
[LG] MiniMax Sparse Attention: MiniMax is a surprisingly effective reward for instruction following
[MiniMax]
https://arxiv.org/abs/2606.13392
---
[LG] Decentralized Multi-Agent Systems with Shared Context: Decentralized Multi-Agent Systems with Shared Context
[Stanford University]
https://arxiv.org/abs/2606.10662
---
[RO] FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning: FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning
[CMU]
https://arxiv.org/abs/2606.12406
---
[CV] MilliVid: Hierarchical Latents for Long-Range Consistency in Video Generation: MilliVid: Hierarchical Latents for Long-Range Consistency in Video Generation
[MIT & Toyota Research Institute]
https://arxiv.org/abs/2606.09056
---
[CL] TimpaTeks: Automatic In-place Text Sequence Modification via Diffusion Language Model Steering: TimpaTeks: Automatic In-place Text Sequence Modification via Diffusion Language Model Steering
[MBZUAI]
https://arxiv.org/abs/2606.08408
外观
