[人人能懂] AI的“灵魂拷问”：当机器开始打草稿、犯糊涂、学做人

00:01:35 AI的“悄悄话”：我们还能“偷听”多久？00:06:10 AI：那个懂所有菜谱，却不会做饭的大厨？00:11:08 AI训练老大难：如何让机器“学徒”少走弯路？00:16:00 给AI动“开心手术”：我们如何让机器更懂“人情世故”？00:19:34 AI的下一个金矿，藏在一只虫子的大脑里？今天介绍的五篇论文：[LG] Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety  [UK AI Security Institute & Apollo Research]  https://arxiv.org/abs/2507.11473 ---[LG] Comprehension Without Competence: Architectural Limits of LLMs in Symbolic Computation and Reasoning  [Amazon Web Service]  https://arxiv.org/abs/2507.106 ---[LG] Relative Entropy Pathwise Policy Optimization  [University of Toronto & Technische Universitat Wien & University of Pennsylvania]  https://arxiv.org/abs/2507.11019 ---[CL] Internal Value Alignment in Large Language Models through Controlled Value Vector Activation  [University of Science and Technology of China & Renmin University of China Beijing]  https://arxiv.org/abs/2507.11316 ---[LG] Biological Processing Units: Leveraging an Insect Connectome to Pioneer Biofidelic Neural Architectures  [Johns Hopkins University]  https://arxiv.org/abs/2507.10951 在小宇宙查看该单集文稿