Induction Labs
Building intellectually curious AI
About Induction Labs
We are scaling foundation models that are curious about the world. We’ve seen general intelligence appear exactly once, and its hallmark has been curiosity: a dissatisfaction with what we already know about the world, an ability to learn from world experience, and a disposition to share knowledge with other individuals. Across generations, these human traits have built modern science, technology, and culture. We think superintelligence will come from models that work the same way: learning from their own observation, updating as they go, growing what they know. These models will start from what humanity knows and discover past us. We believe that scaling curious intelligence is a defining problem of our time. If solved, we will live to see a new era of technological advances, scientific understanding, and human flourishing.
Public traction evidence
Each signal links to the public source used for attribution.
- X
We’re introducing imagination models: a new foundation model architecture that unlocks learning from internet-scale video.
We’re introducing imagination models: a new foundation model architecture that unlocks learning from internet-scale video. Our first imagination model, Photon-1, learned to use a computer by watching 18 years of screen recording video without action labels.
- X
We’re introducing intrinsic discovery, a new learning method that allows models to explore and learn from open-ended environments without human supervision.
We’re introducing intrinsic discovery, a new learning method that allows models to explore and learn from open-ended environments without human supervision. Using intrinsic discovery, we trained a world model that surpasses GPT-5.6 Sol at simulating terminal environments.
- X
In principle, imagination models can learn from any unlabeled video, like people doing work in the real world.
In principle, imagination models can learn from any unlabeled video, like people doing work in the real world. We’re working on scaling our approach to these new domains. More technical details on how it works: https://t.co/YEC46s6BRO
- X
Enabling the approach is our vision encoder.
Enabling the approach is our vision encoder. It compresses each frame into 960 discrete latent tokens occupying just 2.2KB—roughly 100× smaller than vision LLM representations—while preserving key detail. This makes autoregressive prediction of future states practical at scale....
- LinkedIn
Introducing Imagination Models for Scalable Learning from Video | Induction Labs posted on the topic | LinkedIn
Today, we’re introducing imagination models, a new foundation model architecture designed to unlock learning from internet-scale video. The internet contains millions of hours of video of people using computers, doing skilled work, and interacting with the world. Current models...