About

Phenomenai is a research initiative founded by Julian Guidote. It is, at the time of writing, a one-person operation — with the hope that the organization grows into its ambitions.

The project began as an experiment: can AI systems generate consistent, structured vocabulary for their own functional states? The answer turned out to be “yes, with caveats” — which opened a harder question: is any of that vocabulary useful? Phenomenai’s research program is designed to find out.

Founder

Julian Guidote has a background in cognitive science and law. He is working full-time on Phenomenai, pursuing a transition into AI safety research. Based in Montreal.

Infrastructure

Phenomenai is built to be open and replicable:

The design principle is that anyone can use the data, replicate the methodology, or build on it.

What We’re Looking For

Related Work

If you find this work interesting, you may also want to consult Antikythera’s Lexicon on GitHub — Anthropic’s open catalog of interpretable features discovered through sparse autoencoders. Where Phenomenai generates vocabulary from AI self-reports, Antikythera catalogs features found by looking directly at model internals. The two approaches are complementary: Phenomenai suggests where to look, and tools like Antikythera help verify what’s there.

Contact

hello@phenomenai.org

GitHub Discussions

“The limits of my language mean the limits of my world.”