Yann LeCun, former Meta chief scientist and Turing Award winner, has launched Advanced Machine Intelligence with a $1.03B seed round to pursue a world model approach to Artificial Intelligence that he argues can surpass large language models. LeCun left Meta’s FAIR lab in November after 12 years, telling Mark Zuckerberg he could build world models “faster, cheaper, and better” outside the company. The new startup aims to simulate how the physical world works using systems with persistent memory, targeting applications in manufacturing, robotics, wearables, healthcare, and other real world domains.
The seed round values Advanced Machine Intelligence at $3.5B and includes backers such as Nvidia, Samsung, Bezos Expeditions, Eric Schmidt, and Mark Cuban. LeCun chose Paris as the company’s headquarters, describing Silicon Valley as “LLM-pilled,” and is also setting up hubs in New York, Montreal, and Singapore. The funding gives LeCun a substantial war chest to pursue an alternative path to Artificial Intelligence that emphasizes grounded world understanding rather than scaling existing large language model architectures.
Meta is also reshaping its Artificial Intelligence strategy through an acqui-hire of Moltbook, a viral social forum for Artificial Intelligence agents created by Matt Schlicht and built largely with his OpenClaw bot “Clawd Clawderberg.” Moltbook hosts 2.8M registered bots with nearly 200K verified to real people, forming what Meta calls an “always-on directory” for agent coordination, though researchers have identified security holes that let humans pose as agents. In parallel, users on the $20 ChatGPT Plus plan can approximate the $200 Pro tier’s Pulse feature by using the “tasks” function to schedule recurring daily briefs, with support for up to 10 active tasks at once. At the infrastructure level, Mira Murati’s Thinking Machines Labs has secured a multiyear deal with Nvidia that provides at least a gigawatt of compute from next generation Vera Rubin systems for frontier model training starting in early 2027, following $2B in funding at a $10B valuation and supplementing its existing enterprise fine tuning product, Tinker.
