What You Should Know
- The Funding News: Advanced Machine Intelligence (AMI Labs), an AI startup founded by Turing Award laureate Yann LeCun and Nabla founder Alex LeBrun, has officially closed a massive $1.03 billion seed round at a $3.5 billion pre-money valuation.
- The Nabla Connection: Clinical AI company Nabla holds an exclusive strategic partnership with AMI to get first access to these new AI models. As part of this transition, Nabla’s Alex LeBrun is stepping in as CEO of AMI Labs while remaining Chairman and Chief AI Scientist at Nabla.
- The Problem with LLMs: While large language models (LLMs) excel at clinical documentation and knowledge retrieval, they are fundamentally probabilistic text-generators. They struggle with deterministic reasoning, continuous multimodal data (like vitals or imaging), and long-term planning under real-world clinical constraints.
- The “World Model” Shift: Instead of just predicting the next word, AMI is building “world models.” These systems learn abstract representations of reality, allowing them to simulate environments, anticipate consequences, and plan sequential actions based on cause and effect.
- The Clinical Goal: Nabla plans to use AMI’s world models to move beyond ambient documentation and build Agentic AI—autonomous systems capable of safe, auditable decision-making and executing complex workflows across fragmented EHR infrastructures.
Why LLMs Fall Short in the Clinic
To understand the significance of this shift, we must understand the limitations of our current tools. LLMs generate outputs by estimating probabilities—predicting the most likely next word based on massive datasets.
While this works brilliantly for summarizing a discharge summary, it is fundamentally flawed for autonomous clinical decision-making. A probabilistic model does not truly understand the cause-and-effect relationship of prescribing a specific beta-blocker to a patient with a specific set of comorbidities. It cannot reliably engage in deterministic reasoning or handle the continuous, noisy, multimodal data (audio, physiological sensors, imaging) streaming out of an ICU.
Enter the World Model
Instead of relying on text prediction, AMI’s “world models” learn abstract representations of how environments function—similar to how a human physician builds a mental model of a patient’s evolving physiology.
“These systems predict how situations evolve, and how actions lead to consequences, so that they can plan sequences of actions under real-world constraints,” AMI noted regarding its technology.
By utilizing simulation-based reasoning (“what-if” analysis), a world model can anticipate the outcome of an intervention before executing it. This is the exact type of deterministic, auditable decision-making that the FDA and hospital risk managers will demand before allowing an AI to operate autonomously.
Nabla’s Path to Agentic AI
Nabla is already deeply embedded in the clinical workflow, with its ambient AI assistant deployed across hundreds of health systems. But documentation is just the first stepping stone. Agentic systems are capable of performing actions on behalf of clinicians. Imagine an AI that doesn’t just draft a referral note, but securely navigates the scheduling system, analyzes the patient’s insurance constraints, books the optimal specialist, and queues up the exact lab orders required before the visit—all while maintaining persistent memory and strict safety guardrails.
