今天我们来聊聊AI的“内心世界”：我们找到了那把能解锁所有学习方法的“万能钥匙”，却也发现AI的“人格”竟会随着对话见风使舵。我们试图让它像生物一样“进化”，却不小心让它患上了“灾难性遗忘症”。面对越来越强的AI，我们这些“菜鸟裁判”又该如何确保它的诚实？最后，我们会发现，让AI飞速成长的秘诀，可能不是好评，而是一份详尽的“错误报告”。

00:00:32 人工智能的“万能钥匙”藏在哪？

00:06:34 AI的“人格”，为什么聊着聊着就变了？

00:11:47 AI的“进化”陷阱，为什么学得越多，忘得越快？

00:16:47 菜鸟裁判，如何拿捏顶尖高手？

00:21:48 差评，好评，不如一份详细的“错误报告”

本期介绍的几篇论文：

[LG] Spectral Ghost in Representation Learning: from Component Analysis to Self-Supervised Learning

[Google DeepMind & Harvard University]

https://arxiv.org/abs/2601.20154

---

[CL] Linear representations in language models can change dramatically over a conversation

[Google DeepMind]

https://arxiv.org/abs/2601.20834

---

[LG] Evolutionary Strategies lead to Catastrophic Forgetting in LLMs

[UC Berkeley]

https://arxiv.org/abs/2601.20861

---

[LG] Truthfulness Despite Weak Supervision: Evaluating and Training LLMs Using Peer Prediction

[UC Berkeley]

https://arxiv.org/abs/2601.20299

---

[LG] Reinforcement Learning via Self-Distillation

[ETH Zurich]

https://arxiv.org/abs/2601.20802