[人人能懂AI前沿] AI的矛与盾、鸡尾酒与社交圈

你有没有想过，AI在考试中拿高分，靠的究竟是真聪明还是“题海战术”？面对海量信息，AI如何练就像人类一样“速读”和“精读”的本领？为了变得更安全，AI甚至学会了“左右互搏”的自我攻防演练。最新论文还告诉我们，AI不仅能把你的声音“调制”成一杯思想的鸡尾酒，甚至在组队时，也在纠结是找“自己人”还是“局外人”。这期节目，我们就来聊聊AI世界里这些有趣又深刻的进化法则。
00:00:34 AI大模型的高分秘诀，竟然是“题海战术”？
00:06:58 AI的“失忆症”，终于有解药了？
00:12:55 想造最强的盾，得先造最强的矛
00:18:53 你的声音，AI要怎么“翻译”成思想？
00:24:03 AI组队，应该找“自己人”还是“局外人”？
本期介绍的几篇论文：
[CL] Post-training is (Massive) Supervised Learning
[Meta AI & The Hebrew University of Jerusalem]
https://arxiv.org/abs/2606.07527
---
[CL] End-to-End Context Compression at Scale
[New York University & University of Maryland & Princeton University]
https://arxiv.org/abs/2606.09659
---
[CL] Learning to Attack and Defend: Adaptive Red Teaming of Language Models via GRPO
[Microsoft AI Red Team]
https://arxiv.org/abs/2606.09701
---
[CL] Is Text All You Need? Text as a Universal Information Bottleneck for Speech LLMs
[Microsoft Research & Microsoft Research Asia & The Chinese University of Hong Kong]
https://arxiv.org/abs/2606.09366
---
[CL] Representational Similarity and Model Behavior in Multi-Agent Interaction
[UC Berkeley & The University of Chicago]
https://arxiv.org/abs/2606.07818