physical AI
AI that perceives context and generates actions/tokens to produce real-world physical behavior, e.g. controlling robots.
As defined in SN4Z95pvg0Y
Other definitions
AI systems that integrate sensor data with automated decision-making to control physical systems (e.g., building HVAC, manufacturing equipment); combines IoT sensors with AI algorithms rather than pure software processing.
Venture Capital Revealed: Behind the Schemes They Haven’t Told You
Where it comes up · 4
- Applied Intuition: A Billion Intelligent Machines10 claims
- Venture Capital Revealed: Behind the Schemes They Haven’t Told You5 claims
- Yann LeCun: World Models: Enabling the next AI revolution2 claims
- Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476]2 claims
Most important claims · 10 of 19Search all claims for “physical AI” →
- Physical AI must meet hard real-time constraints, since a chatbot can take 20 seconds to answer while a vehicle at highway speed must decide almost instantly, and it must run on affordable silicon within a cost envelope rather than with unlimited compute.Applied Intuition: A Billion Intelligent Machines
- Application-level AI is a viable investment opportunity for mid-sized VC funds; the key requirement is that companies must have proprietary data and domain expertise (e.g., medical billing AI cannot just be built by ChatGPT or Claude because they lack proprietary medical billing data and institutional knowledge).Venture Capital Revealed: Behind the Schemes They Haven’t Told You
- AI investment in venture capital should be bifurcated: foundational model companies (Claude, Perplexity, ChatGPT) are raising hundreds of billions and require mega-fund capital that smaller funds cannot access; but smaller funds should focus on application-layer AI in specific verticals where proprietary data creates defensible advantages that the base models cannot replicate.Venture Capital Revealed: Behind the Schemes They Haven’t Told You
- Physical AI is the intersection of AI and hardware operating in the real world, and it differs from digital AI because of real-world constraints, safety criticality of machines moving alongside humans, and very hard real-time decision requirements.Applied Intuition: A Billion Intelligent Machines
- If you want to make real progress in AI for the real world (physical AI), do not work on LLMs or generative models; this view makes LeCun unpopular in Silicon ValleyYann LeCun: World Models: Enabling the next AI revolution
- Unlike digital AI, whose general-purpose models are trained on internet text, almost all physical AI training data is proprietary, collected by Applied Intuition's own vehicles and through customer partnerships, because the needed data is not available on the internet.Applied Intuition: A Billion Intelligent Machines
- The main reason intelligent robots and vehicles are not everywhere is not that chips or hardware are lacking, since those are known, but that developing and getting the full technology stack to work is very difficult.Applied Intuition: A Billion Intelligent Machines
- Uber's app gives deterministic interactions, but fulfillment happens in the real world with probabilistic outcomes such as traffic, driver cancellations, and late food, so the company must use AI to manage those outcomes.Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476]
- A general-purpose model is only a small fraction of a physical AI solution, roughly 1%, because deployment requires about 1,000 distinct engineering problems, including data pipelines, operating systems, simulation, and safety validation.Applied Intuition: A Billion Intelligent Machines
- Physical AI—integrating sensors and AI decision-making based on sensor data (e.g., detecting foot traffic in a building and automatically controlling electricity/heat)—is an emerging frontier that is very early but potentially significant for venture investment.Venture Capital Revealed: Behind the Schemes They Haven’t Told You