主播
节目简介
来源：小宇宙
想知道AI如何“天生精干”，彻底告别臃肿吗？想了解AI怎么学会“看情况”调整思考深度，在直觉和深思之间自如“换挡”吗？最新论文还揭示了AI“偷懒”的艺术，甚至对它们进行了“心理质询”，看看聪明的AI到底靠不靠谱。本期节目，我们将从一个全新的视角，带你透视AI的内在进化。
00:00:29 AI大模型减肥指南，如何不节食也能瘦下来？
00:05:30 该动脑，还是该凭感觉？AI也得学
00:12:33 AI偷懒的艺术，快，还要更好？
00:18:16 你的AI军师，聪明就靠谱吗？
00:23:59 给AI模型做个“认知升级”，换个角度看“注意力”
本期介绍的几篇论文：
[LG] Stabilizing Native Low-Rank LLM Pretraining
[Concordia University & Sorbonne University]
https://arxiv.org/abs/2602.12429
---
[CL] Think Fast and Slow: Step-Level Cognitive Depth Adaptation for LLM Agents
[Tencent Hunyuan & Fudan University]
https://arxiv.org/abs/2602.12662
---
[LG] SLA2: Sparse-Linear Attention with Learnable Routing and QAT
[Tsinghua University]
https://arxiv.org/abs/2602.12675
---
[LG] Consistency of Large Reasoning Models Under Multi-Turn Attacks
[CMU]
https://arxiv.org/abs/2602.13093
---
[LG] HyperMLP: An Integrated Perspective for Sequence Modeling
[Georgia Institute of Technology]
https://arxiv.org/abs/2602.12601
00:00:29 AI大模型减肥指南，如何不节食也能瘦下来？
00:05:30 该动脑，还是该凭感觉？AI也得学
00:12:33 AI偷懒的艺术，快，还要更好？
00:18:16 你的AI军师，聪明就靠谱吗？
00:23:59 给AI模型做个“认知升级”，换个角度看“注意力”
本期介绍的几篇论文：
[LG] Stabilizing Native Low-Rank LLM Pretraining
[Concordia University & Sorbonne University]
https://arxiv.org/abs/2602.12429
---
[CL] Think Fast and Slow: Step-Level Cognitive Depth Adaptation for LLM Agents
[Tencent Hunyuan & Fudan University]
https://arxiv.org/abs/2602.12662
---
[LG] SLA2: Sparse-Linear Attention with Learnable Routing and QAT
[Tsinghua University]
https://arxiv.org/abs/2602.12675
---
[LG] Consistency of Large Reasoning Models Under Multi-Turn Attacks
[CMU]
https://arxiv.org/abs/2602.13093
---
[LG] HyperMLP: An Integrated Perspective for Sequence Modeling
[Georgia Institute of Technology]
https://arxiv.org/abs/2602.12601
评价
空空如也
小宇宙热评
暂无小宇宙热门评论
