Yann LeCun's new venture is a contrarian bet against large language models
3 months ago
- #AI Research
- #Open Source
- #World Models
- Yann LeCun, a Turing Award recipient and AI pioneer, has launched a new Paris-based company, AMI Labs, focusing on world models as an alternative to large language models (LLMs).
- LeCun criticizes the current industry obsession with LLMs, arguing they lack a true understanding of the real world and cannot achieve human-level intelligence without major conceptual breakthroughs.
- He advocates for open-source AI, contrasting with the closed approaches of companies like OpenAI and Anthropic, and sees a demand for a credible, non-US, non-China AI company.
- AMI Labs will focus on JEPA (joint embedding predictive architecture), a framework for AI to learn abstract representations of the world from video, audio, and sensor data, enabling better reasoning and planning.
- Potential applications of world models include industrial processes, smart glasses, domestic robots, and autonomous driving, where predicting real-world dynamics is crucial.
- LeCun left Meta, where he founded FAIR, due to disagreements over strategic directions, but maintains no bad blood and sees potential collaboration with Meta as AMI's first client.
- He believes academia should avoid LLM research, which is now industry-dominated, and instead focus on foundational breakthroughs like world models.
- AMI Labs is global, with offices planned in Paris, North America (likely New York), and possibly Singapore, attracting talent from top AI labs like OpenAI and Google DeepMind.
- LeCun will serve as executive chairman, focusing on science and vision, while Alex LeBrun, a serial AI entrepreneur, will be CEO, handling day-to-day operations.