The post points to Book Prize Index, a free web app that aggregates prize-winning nonfiction and adds semantic search so readers can browse for serious books without relying on generic bestseller lists or chatbot summaries. The creator says the point is not to have AI write books, but to use modern tooling to recover something closer to library browsing and serendipitous discovery. That landed well. People quickly found books they wanted to read, asked for features like author search, more international awards, Libby integration, and better sorting, and flagged bugs in filters, links, and search behavior.
The bigger conversation was about where AI actually helps. The strongest practical consensus was that semantic search over a human-curated corpus is a good use of
LLM-era tooling, while asking models to generate prose or compress books into takeaways is exactly how you get slop. Several people said LLMs are useful as pointers to books, blogs, and obscure sources that normal search no longer surfaces. They were much less impressed by AI as a writer. The common complaint was not just factual shakiness, but that model prose collapses toward average phrasing and loses the intent, friction, and lived specificity that make strong nonfiction or fiction worth reading in the first place.
That bled into a long side argument over whether human creativity is just recombination. The useful landing point was narrower than the metaphysics: even if human creativity also recombines inputs, people have persistent memory, embodied experience, social context, and stakes in the real world that current language models do not. Commenters kept returning to the same distinction. Models can produce plausible descriptions. They do not have firsthand experience, and that gap shows up most when writing has to carry perception, judgment, and voice rather than just surface pattern matching.
A second thread pushed on learning rather than writing. Some readers said long-form books force active integration of ideas in a way chatbot interaction often does not. LLM answers feel easy to absorb and just as easy to forget. Others replied that this is partly about usage. If you use a model as a
Socratic tutor, write down your own understanding, and test yourself, it can support active learning rather than replace it. The practical split was clear: books remain better for slow
schema-changing understanding, while LLMs shine as interactive scaffolding around that process.
There was also healthy skepticism about the site’s core ranking signal. Prize lists are a decent filter, not a truth machine. Publishers spam submissions, some awards are weak or gamed, and awards reflect judge bias, prestige politics, and the tastes of a moment. That skepticism did not kill interest in the project. It just reframed the index as a useful starting point rather than an objective canon.
resobscura.substack.com