Michael Darius Eastwood · Independent AI alignment researcher
Published
This programme has no journal publications. That is a real limitation and the record does not pretend otherwise. But the objection is usually deployed as if peer review were the only mechanism by which claims earn credibility, and that is false. It is worth being precise about what each mechanism buys.
Two or three qualified readers check a paper's methods and reasoning before publication. It is a filter against obvious error and a signal of community standards. It is not a replication, not an audit of underlying data, and, as every retraction scandal demonstrates, not a guarantee.
Three things, each checkable without trusting anyone. First, cryptographic timestamps: the manuscripts have published hashes and downloadable files, so the when of every claim is verifiable by a stranger in minutes. Second, independent convergence: nineteen sourced events, from Anthropic's experiment to a formal undecidability proof, reached the programme's structural conclusions without citing it or knowing of it. Reviewers can be networked to an author; strangers who have never heard of you cannot be. Third, adversarial self-exposure: published kill-conditions for every claim, one retraction already on the record, and a truth gate that lints the corpus like code.
Convergence shows the conclusions are shared; it does not check the programme's own experiments, which remain single-lab and are flagged as such on every claim they support. That is why the current work is a replication package anyone can run and an arXiv submission that invites exactly the scrutiny this objection asks for. The honest position is not that peer review is unnecessary. It is that verification does not have to wait for it.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.