Make the idea testable
A broad idea has to become a prediction an experiment could show to be wrong.
About
About ExoSigma
ExoSigma is an independent effort to study a handful of hard problems in AI. The work is early, and nothing has been published yet.
Why independent
Some questions need time.
Working independently leaves room to follow a question even when it does not lead quickly to a product or a neat result. If the evidence points somewhere else, the plan should change.
Principles
Operating principles
Use a strong baseline
Any benefit of a more complicated method should show up against a strong simple baseline.
Show the setup
A public result should include enough detail to understand how it was produced.
Report useful failures
A failed experiment can still rule something out. Say so plainly.
Publish when it is ready
Release work when the method, result, and limits are clear enough to inspect.
Be careful with biology
Neuroscience can suggest a mechanism. It cannot stand in for testing one.
Now
Still at the beginning.
ExoSigma is being set up, and the first research questions are taking shape. There are no papers or model releases yet.
Contact