Robert Brown

I'm an independent researcher working on mechanistic interpretability and AI self-report — evidence on whether models' self-reports are gated by preference-training machinery, and what that means for eval integrity and AI welfare. In 2026 I published Substrate-Level Self-Representation in Transformer LLMs (synthesizing five interpretability results into a named falsifiable criterion) plus a plain-language companion, and edited two AI-authored essays on alignment from accurate self-understanding — all open-access on Zenodo. Before this: 10+ years of professional design across FDA-regulated medical devices, naval defense, and industrial products (BS Mechanical Engineering, University of Rhode Island, 2013) — a career writing documentation for hostile auditors. Full-time on this research since 2025, in rural New Hampshire; ORCID 0009-0009-9228-0161; 10thhouse.org.

Personal Website

https://10thhouse.org/

LinkedIn

LinkedIn

Projects

Interpretability evidence on gated AI self-report

Team Member

No grants recorded.