AI Agent Produces First QFT Textbook Draft
On his 'Not Even Wrong' blog, mathematician Peter Woit reports, based on a Twitter thread he cites, that physicist Xi Yin is supervising an effort to have an AI model (described in the thread as GPT-5.5) generate a multi-volume quantum field theory textbook. Woit points to a public GitHub repository, xiyin137/QFT, whose stated workflow has contributions reviewed and approved by Xi Yin; Woit says the version he compiled produced a roughly 3,500-page PDF and that he saw very recent commits. He characterizes the draft as "clearly a work in progress," unusable as a conventional textbook but possibly of limited use to experts. Woit also relays unconfirmed rumors, discussed in the same thread, that Xi Yin may have been hired by OpenAI. The episode is a concrete, if early, look at agent-generated long-form technical writing.
What it is
On his long-running 'Not Even Wrong' blog, Columbia mathematician Peter Woit describes an experiment, which he learned of from a Twitter thread, in which physicist Xi Yin is supervising the use of an AI model, identified in the thread as GPT-5.5, to draft a multi-volume quantum field theory textbook. The work lives in a public GitHub repository, xiyin137/QFT.
What Woit observed
Woit says he cloned the repository and compiled the LaTeX source into a PDF that ran to roughly 3,500 pages, and that the repository showed very recent commit activity. He describes the output as "clearly a work in progress," not usable as a conventional textbook, though potentially of narrow use to experts who can already judge the material.
Caveats
Several details, including the specific model and a rumor that Xi Yin may have joined OpenAI, come from the cited thread rather than confirmed reporting, and should be treated as such. The repository's own workflow notes that contributions are reviewed by Xi Yin, which suggests human oversight of the agent's output.
Why it matters
For researchers and tool builders, the value is as a concrete data point on agent-generated long-form technical writing: models can now emit enormous, structured, compilable documents, but correctness and pedagogy still require human curation. Public repositories make it possible to fork, review, and gradually vet such drafts, hinting at a workflow from raw generation to trusted resource.
Key Points
- 1Agent-generated monographs can produce large volumes of technical text fast, but page count is not pedagogical quality, especially for rigorous subjects like QFT.
- 2Publishing output as LaTeX in a public, version-controlled repo enables community review and tooling, yet mathematical correctness still demands expert verification.
- 3The project is a single-researcher, work-in-progress experiment reported via blog, useful as a signal about agent capability rather than a finished resource.
Scoring Rationale
A concrete demonstration of an AI agent producing a very large, compilable technical draft is genuinely interesting to researchers and tool builders, but this is a single-researcher, work-in-progress experiment reported via a personal blog and an unverified thread, with limited immediate utility. Scored as a solid, niche signal about agent capability rather than a major research result.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems