LeCun 对 LLM 的批判性观点和世界模型路线图，对 AI 研究者、博士生和关注下一代 AI 架构的人极具启发，值得花一小时听完。
Yann LeCun 在 Unsupervised Learning 播客中与 Jacob Effron 进行了深度对话，分享了他对 LLM 局限性的尖锐观点，以及为何与 Hinton、Bengio 在 LLM 问题上产生重大分歧。他透露了离开 Meta 的原因，并介绍了新公司 AMI 押注世界模型的战略。LeCun 还预测了 2027 年的 AI 发展，并建议博士生不要再专注于 LLM 研究。这场访谈涵盖安全讨论、FAIR 的得失以及突破性研究如何真正发生。
Fun interview with Jacob Effron on the Unsupervised Learning podcast.
Fun interview with Jacob Effron on the Unsupervised Learning podcast. Jacob Effron @jacobeffron It’s hard to imagine more of a dream Unsupervised Learning guest than @ylecun . Yann is one of the godfathers of AI, and he has some fascinating contrarian views on the limitations of LLMs. It was incredible to get to have a wide-ranging discussion with Yann about these views, reflections on his time at Meta and departure and what’s next for him. We hit on: ▪️ LLM limitations and a path forward for robotics ▪️ Why he left Meta ▪️ How he came to so dramatically disagree with his Turing co-laureates Geoff Hinton and Yoshua Bengio on LLMs ▪️ His predictions for 2027 ▪️ His new company AMI and the bet on world models ▪️ Why he compares OpenAI and Anthropic to Sun Microsystems ▪️ Why he tells PhD students to stop working on LLMs Plus some sharp views on the current safety discourse, how breakthrough research actually happens and what FAIR got right and wrong. YouTube: youtu.be/ngBraLDqzdI Spotify: bit.ly/4dL8fvT Apple: bit.ly/4wxgpiX Your browser does not support the video tag. 🔗 View on Twitter 🔗 View Quoted Tweet 💬 24 🔄 48 ❤️ 388 👀 54421 📊 71 ⚡