Archetype AI is a foundational AI company bringing intelligence to the physical world.
Visit Company
Why We Invested in Archetype
Aug 2023
Archetype is building a multimodal foundational model that unlocks a whole new paradigm for AI, through the physical world. This is game changing — this means any company building anything of value in the physical world can use Archetype’s models to build intelligence on top of their sensors.
For context, each modern automobile can contain more than 200 sensors. Large-scale industrials like factories or power grids can contain tens of thousands, if not millions of sensors.
At present, most if not all LLMs (large language models) are trained on text and the semantic web. Archetype takes on a differentiated approach with their multimodal foundational model, which takes in any type of sensor input and encodes them into a universal embedding space. Beyond just text and images, Archetype’s models take in time-series data from sensors such as radars, cameras, accelerometers, gyroscopes, and encodes them into a unified representation of the physical world.
They call their first Large Behavior Model, Newton. Behavior is a reference to the time series data that tracks changes or trends over time, and the insights that come from patterns and trends observed or derived.
An API for the Physical World
This is what makes the Archetype team so special— beyond being brilliant scientists and developers, they also take a user-centered approach to their designs. With Newton, they make it not only possible to compute multimodal sensor data, but then enable users to decode it in any way, be it via text query, data visualizations, or code. Imagine the power of chatGPT integrated with data from the physical world.
By integrating Newton, developers and enterprises can summarize, search, predict, and intelligently monitor sensor data, and even perform actions like optimizing or automating tasks based on inferences. A similar analog could be drawn to the instruct models that truly unleashed GPT for the mainstream user, by enabling them to interact with the intelligence models in a natural, query-and-response way.
The interaction paradigms will make it fundamentally easier for developers and enterprises to build useful, and delightful, physical AI applications for the end-user. Use-cases range from manufacturing and logistics tracking, robotics, consumer electronics. Because Newton, their Large Behavior Model, is designed to work with an extensive variety of inputs—from large-scale industrials to pico motion sensors—it enables intelligence at every scale.
A Tight-Knit Team Who Have Shipped Inventions Together
The cofounders of Archetype — Ivan Poupyrev, Brandon Barbello, Leonardo Giusti, Nick Gillian, and Jamie Lien — worked on several patented inventions together while at Google for the past 5 years. As I’ve gotten to know the team over the past year, it has further reinforced my observations that the team has incredible synergy. Many of them are cross-disciplinary and are able to build upon (or sometimes, challenge) each other’s ideas and concepts, which makes it an extremely effective and creative environment for pushing the frontiers.
The team has grown a strong developer following and community over the past few decades, from their work building the Soli Radar to their individual inventions at previous companies like Sony, Disney, MIT Media Lab, and more.
Archetype’s developer-first approach, both in building Newton and their community, points towards the potential of a strong developer community that’s attractive to both enterprises as well as the independent, open source community.
If you’re not following Archetype yet, you should. If you’re an enterprise ready to explore a pilot partnership, please reach out to me or info@archetypeai.io. They have a long waitlist, but will prioritize strategic use-cases. Developer access is coming soon. Sign up on their website to be notified.
