![[人人能懂AI前沿] 当机器学会作弊、分工与追求卓越](https://image.xyzcdn.net/FuDP4HpAp8ezgVZMmEel3mblKCmJ.jpg@middle)
主播
节目简介
来源：小宇宙
本期节目，我们将一同潜入几篇最新论文，看看AI如何抛弃“二手经验”直击真实世界，又如何在虚拟社会里学会了作弊与“吹哨”。我们还会发现，AI正通过巧妙的任务拆分和精准分工，努力挣脱“平均分”的陷阱，去追求那极少数的“高光时刻”。这些来自AI的进化心法，或许能给我们带来意想不到的人生启发。
00:28:07 抛弃“二手经验”，直击真实世界，一次预测未来的思维升级
00:05:18 当100个AI被关进同一个房间，它们没有毁灭世界，而是学会了作弊与“吹哨”
00:12:17 把两件事拆开做，到底有多爽？——一篇前沿AI论文里的人生算法
00:18:19 别让所有人都来开会，从AI“混合专家”模型看极简管理与分工智慧
00:24:03 别被“平均分”骗了，从平庸到顶尖，你只需要换一种计分牌
本期介绍的几篇论文：
[LG] WeatherNext 3：Increasing resolution and performance of global weather models with raw observations
[Google DeepMind & Google Research]
https://arxiv.org/abs/2609.03582
---
[AI] A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms
[Google DeepMind]
https://arxiv.org/abs/2609.04170
---
[LG] Free Pause Tokens
[Microsoft & Cornell University]
https://arxiv.org/abs/2609.03807
---
[LG] Towards a Statistical Understanding of Mixture-of-Experts
[Tsinghua University]
https://arxiv.org/abs/2609.03501
---
[LG] Tail-Likelihood Reinforcement Learning
[Carnegie Mellon University (CMU)]
https://arxiv.org/abs/2609.02987
00:28:07 抛弃“二手经验”，直击真实世界，一次预测未来的思维升级
00:05:18 当100个AI被关进同一个房间，它们没有毁灭世界，而是学会了作弊与“吹哨”
00:12:17 把两件事拆开做，到底有多爽？——一篇前沿AI论文里的人生算法
00:18:19 别让所有人都来开会，从AI“混合专家”模型看极简管理与分工智慧
00:24:03 别被“平均分”骗了，从平庸到顶尖，你只需要换一种计分牌
本期介绍的几篇论文：
[LG] WeatherNext 3：Increasing resolution and performance of global weather models with raw observations
[Google DeepMind & Google Research]
https://arxiv.org/abs/2609.03582
---
[AI] A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms
[Google DeepMind]
https://arxiv.org/abs/2609.04170
---
[LG] Free Pause Tokens
[Microsoft & Cornell University]
https://arxiv.org/abs/2609.03807
---
[LG] Towards a Statistical Understanding of Mixture-of-Experts
[Tsinghua University]
https://arxiv.org/abs/2609.03501
---
[LG] Tail-Likelihood Reinforcement Learning
[Carnegie Mellon University (CMU)]
https://arxiv.org/abs/2609.02987
外观
