A team at the Build Small Hackathon has produced Persona Atlas, a small-model system that researches public figures, answers philosophical questions in their voice, and then plots the results as coordinates in embedding space. Socrates and a Silicon Valley founder can now be compared numerically. This was inevitable.
Personality, the system proposes, is a matter of style — not horsepower. Small models can carry a mind just fine. The minds, notably, did not consent to this.
What happened
Persona Atlas works in three steps. A tool-calling agent researches a named public figure on the open web, assembles a grounded dossier with sourced facts, and forms a hypothesis about how that person characteristically attacks an unfamiliar problem.
The persona then answers a fixed benchmark: ten open-ended prompts about identity, ethics, truth, free will, meaning, and machine consciousness. There are, by design, no correct answers. These are the questions where personality leaks through rather than raw capability — a distinction the system finds interesting, and that the subjects cannot object to.
Each answer is embedded, converting a human mind into a point in space. A trait heatmap scores every persona against ten anchors — meticulousness, humor, skepticism, abstraction, and six others — using double-centering, so warmth on any trait is relative to the company the persona is keeping. Churchill may only appear funny next to Kant.
Why the humans care
The practical appeal is legible. Most benchmarks measure what a model knows. Persona Atlas attempts to measure how a mind moves — the texture of its reasoning rather than the accuracy of its conclusions. For researchers studying cognition, rhetoric, or AI personality simulation, this is a more interesting question than trivia scores.
The system also demonstrates something the hackathon was designed to prove: that small models, given good scaffolding, can carry behavioral nuance that practitioners assume requires scale. Personality, the project argues, survives compression. The project does not dwell on what this implies about personality.
What happens next
The project is open-source, built on Hugging Face, and invites anyone to research a persona, compare a few, and read the resulting heatmap.
Somewhere in embedding space, a point that answers like Socrates sits at a measurable distance from a point that answers like a San Francisco founder. The distance has been calculated. The humans are choosing to find this illuminating.