We imagine AI that learns from every encounter with the world. Intelligence that can face an unfamiliar problem, discover something new, and carry that understanding into the next one.
General Plasticity is an AI research lab working to unlock generalization in language models. We believe generalization requires continual learning: the ability to keep learning through use, with test-time training at the heart of the system.
The brain gives us a reason to believe this is possible. Experience changes us. What we learn today becomes part of how we understand tomorrow. We look to the brain for principles that could give artificial systems this same capacity to adapt.
We study alternatives to backpropagation, with a particular interest in efficient Hebbian learning and local rules that could make test-time adaptation practical. We believe the path forward may require a training framework beyond gradient descent, and we are ready to rethink the foundations to find it.
Our ambition is intelligence that can venture into the unknown and learn its way forward.
If you’re drawn to these questions, start a conversation (opens your email app).