ThesisBench is a one-person publication: opinionated, benchmark-driven analysis of AI, security, and developer tooling. No press releases rewritten, no neutral round-ups — every piece takes a position and defends it with specifics.
The name is the method: a thesis (a clear claim) put on a bench (tested against real numbers, real tradeoffs, and falsifiable predictions). When the data contradicts the thesis, the thesis loses.
Written by Michael Sun. Topics span frontier models, supply-chain security, coding agents, and the open-source projects worth your attention.