![[人人能懂AI前沿] 从边想边画、价值渗漏，到机器人的“脑内沙盘”](https://image.xyzcdn.net/FuDP4HpAp8ezgVZMmEel3mblKCmJ.jpg@middle)
主播
节目简介
来源：小宇宙
今天我们要分享五篇极其有趣的最新论文，带你看看AI不仅学会了像人类一样“边思考边画画”并随时纠错，竟然还长出了带有偏见和私心的“小算盘”。此外，我们还要探究大模型总是“学了新知识就忘旧知识”的失联真相，并尝试给笨手笨脚的机器人戴上一副看懂物理空间的“母语眼镜”。最后，我们一起看看科学家如何通过“脑内推演”和“跨界对齐”，让机器人彻底告别“一根筋”。准备好刷新你对人工智能的认知了吗，我们马上出发！
00:00:37 AI的新活法，一边思考，一边画画
00:05:56 你以为AI是中立的？其实它有自己的“小算盘”
00:11:41 给机器人换一副“母语”眼镜
00:17:44 给AI上课，为什么它学会了新知识，却忘了旧的？
00:24:41 让机器人告别“一根筋”
本期介绍的几篇论文：
[LG] Concurrent Image Understanding and Generation: Self-Correcting Coupled Markov Jump Processes
[Google]
https://arxiv.org/abs/2607.13188
---
[LG] Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values
[Truthful AI]
https://arxiv.org/abs/2607.14345
---
[RO] See like a Robot: Robot-Centric Pointmaps for Vision-Language-Action Models
[KAIST AI]
https://arxiv.org/abs/2607.11498
---
[CL] Can a Language Model Learn Facts Continually in Its Weights?
[Baseten]
https://arxiv.org/abs/2607.11020
---
[RO] Towards Predictive, Aligned, and Scalable Robot Learning
[Astribot Team]
https://arxiv.org/abs/2607.11270
00:00:37 AI的新活法，一边思考，一边画画
00:05:56 你以为AI是中立的？其实它有自己的“小算盘”
00:11:41 给机器人换一副“母语”眼镜
00:17:44 给AI上课，为什么它学会了新知识，却忘了旧的？
00:24:41 让机器人告别“一根筋”
本期介绍的几篇论文：
[LG] Concurrent Image Understanding and Generation: Self-Correcting Coupled Markov Jump Processes
[Google]
https://arxiv.org/abs/2607.13188
---
[LG] Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values
[Truthful AI]
https://arxiv.org/abs/2607.14345
---
[RO] See like a Robot: Robot-Centric Pointmaps for Vision-Language-Action Models
[KAIST AI]
https://arxiv.org/abs/2607.11498
---
[CL] Can a Language Model Learn Facts Continually in Its Weights?
[Baseten]
https://arxiv.org/abs/2607.11020
---
[RO] Towards Predictive, Aligned, and Scalable Robot Learning
[Astribot Team]
https://arxiv.org/abs/2607.11270
外观
