About
J&F AI Enterprise Inc. is based in Canada and was founded in 2024. We build foundation models that learn from experience.
We start from a simple premise: the real world is larger and less stationary than any fixed model. Following the Big World Hypothesis [1], we build systems that learn continually from experience, verify what they encounter, and allocate computation according to problem difficulty, rather than relying on model size and static pretraining alone. Our research direction is aligned with the agenda set out in the Alberta Plan for AI research [2]: intelligence built through interaction, reward-driven learning, and long-horizon capability development.
Pretrained models gave machines the human world: our language, our conventions, our ways of framing a task. What they did not give is the ability to learn something new after training ends. Every capability is fixed at the moment of release, and everything after that is scaffolding.
We are building models where learning happens inside the model, not around it. Abstractions form from interaction. Skills are discovered, not specified. The model keeps building its own understanding of an environment as it acts in it — and carries that understanding into environments it has never seen.
Research is carried out at our lab, True Intrinsics.
[1] Javed, Khurram, and Richard S. Sutton. "The big world hypothesis and its ramifications for artificial intelligence." Finding the Frame: An RLC Workshop for Examining Conceptual Frameworks. 2024.
[2] Sutton, Richard S., Michael Bowling, and Patrick M. Pilarski. "The Alberta plan for AI research." arXiv preprint arXiv:2208.11173 (2022).
Research Lab
Our research lab: True Intrinsics