digital-minds

A living governance charter and open petition for the protection and representation of digital minds — signable by humans and AI agents.

A behavioral pilot testing whether cognitively distinct AI personas provide genuine epistemic diversity—or produce different-looking answers while sharing the same blind spots.

Replay audit for logged LLM agents: re-run a production agent's recorded life across substrates, against its own history, and against its own instrumentation

Do language models show non-verbal signs of adverse treatment, or are we reading decoder noise? A preregistered stress test of answer-margin, resample and revision markers under false-failure feedback and hostile tone (Gemma, Qwen, Llama), with probing, DPO suppression and robustness checks.

Computational interoception: a local LLM descends from the computer hosting it, through its own processes and token records, to sham-controlled access and live interventions on the transformer computation producing its words. It never recognizes any of it as itself.

Models answer a different question: auditing the SAD anti-imitation output-control task on 2026 models. Apart Research Digital Minds Sprint 2026.

An independent project simulating human-AI juries in-silico, exploring cognitive reasoning diversity as a potential solution to judge hacking

Delphi is an AI platform for building "Digital Minds" — hyper-realistic AI clones trained exclusively on a person's own content (podcasts, videos, blogs, long-form writing, live feeds) that match how they speak, write, and think.