At the unsettled edges of AI.
ExoSigma is an AI research lab. We work on directions the field has set aside, drawing hypotheses from how the brain generalizes.
Where the field is not looking.
Most of the field is moving along one path: larger models, more data, and refinements to a narrow set of architectures. That path has been extraordinarily productive, and it has absorbed most of the field’s attention. Directions that lost out along the way — ideas tried at a scale or in a setting that no longer applies, or never given a serious attempt — are rarely revisited. We think that is where the most interesting open ground now sits.
Generalization, by way of the brain.
Our starting point is generalization. The brain learns from little experience and carries what it learns into situations it has never seen, and neuroscience now describes pieces of how it does so in computational terms. We take those descriptions as hypotheses about what a learning system could do differently, and we pursue the ones that point away from current practice rather than toward it.
A lab built to try more of them.
We are building ExoSigma from the ground up as an agent-run research pipeline. Agents survey the literature, vet candidate ideas against what has already been tried, design and create experiments, and run the automation around them. We set the direction and decide what the results mean. The point is throughput of ideas: to give many set-aside directions a fair attempt, not just a few.