Alex LeBrun's Recent LinkedIn Posts

We're building one of the most ambitious AI company in the world. So naturally, my most urgent problem is: I have no idea who's doing payroll. [1] We're looking for an 𝐇𝐑 𝐎𝐩𝐬 to join AMI Labs in Paris and build our people function from scratch — across Paris, Montreal, New York, and Singapore. You'll own everything: onboarding, contracts, visas, payroll (in 4 countries 🙃), benefits, performance cycles, HR systems. The full stack, but for humans. The ideal profile: 7-10 years of HR ops experience, startup DNA, comfortable being the entire HR ops department for a while — and somehow still excited about that. Bonus points if you've worked in a research lab before. Double bonus if you don't panic when I suddenly ask "wait, do we have a parental leave policy?". 📍 Paris (in person at the AMI office) [1] Actually I do know. I hope.

I am thrilled to announce that I am joining Yann LeCun and an exceptional founding team to lead AMI - Advanced Machine Intelligence as CEO. We have secured a $1.03 billion USD (approximately €890 million) seed round to fuel our mission to build intelligent systems capable of truly understanding the real world—a long-term scientific endeavor. Having spent 24 years building AI products—including work on chatbots (VirtuOz, 2002), natural language understanding (Wit.ai, 2012), Facebook's M assistant (2015), and healthcare AI (Nabla, 2019)—I have witnessed the strengths and, crucially, the limitations of machine learning methodologies. While the solutions we've shipped have (I hope) had a positive impact and helped people, they all relied on "shortcuts". Generative architecture trained by self-supervised learning 𝑚𝑖𝑚𝑖𝑐 intelligence; they don't genuinely understand the world. Predicting tokens, though powerful, works best for discrete and low-dimensional tasks like information retrieval, summarization, coding, and mathematics. However, factories, hospitals, and robots operating in open environments demand AI that grasps reality. And reality is not tokenized: it’s continuous, noisy and high-dimensional. Despite their immense power, I do not believe that generative architectures are the path to achieving this true understanding. It is time to move beyond shortcuts and work on a foundational solution. 𝐖𝐨𝐫𝐥𝐝 𝐦𝐨𝐝𝐞𝐥𝐬 learn abstract representations of real-world data, ignoring unpredictable details, and make predictions in representation space. Action-conditioned world models allow agentic systems to predict the consequences of their actions, and to plan action sequences to accomplish a task, subject to safety guardrails. AMI will advance AI research and develop applications where reliability, controllability, and safety really matter, especially for industrial process control, automation, wearable devices, robotics, healthcare, and beyond. Nabla is AMI’s first partner. Nabla's customers will gain early access to our next-generation world models research, which will complement LLMs, strengthening our strategy to build safe, reliable, and clinically meaningful AI systems for healthcare. I will be transitioning from CEO of Nabla to Chief AI Scientist and Chairman, continuing to work closely with my co-founders Martin Raison Delphine Groll Laurent Landowski and the incredible team that has already made a profound impact on millions of physicians and patients. If AMI’s mission resonates with you, we invite you to join us on this journey! 𝐑𝐞𝐚𝐥 𝐰𝐨𝐫𝐥𝐝. 𝐑𝐞𝐚𝐥 𝐢𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞.

My friend: "LLMs are perfect, we don't need World Models." LLMs answering "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?" 👇 [Credit: Kévin @knowmadd@mastodon.world]