After Karpathy's HackerTube, a clone mines 15k books from Hacker News into HackerBooks
huangyun_122 · x · 2026-11-03
Following Karpathy's HackerTube (built in two hours with Opus 5.5 on HuggingFace's public HN dataset), an imitator built HackerBooks, extracting 15,000+ books from Hacker News posts (each mentioned by at least two people) — topped by classics like TAOCP and GEB.
Blogger huangyun122 tested Grok on the list: asked to pick a CS book it hadn't read, it chose the 2025 title Building a Debugger and claimed to know only 66 of them. He raises a broader question: why don't LLM pretraining corpora include these curated collections — joking that's probably why GPU prices keep rising.
Related event: After Karpathy's HackerTube, Copycats Build HackerBooks from HN Data(2 posts)→
More from coding & agent
- 'The order is the control': anchoring-proof agent pipeline puts hard rules before the model — Neither_Opinion69 · 2026-11-03
- Line counter loccount hit 803 languages — LLMs added 31 in two days, double the runner-up — esrtweet · 2026-11-03
- Autoresearch agents get stuck in local optima — the fix may be making them forget — eshorten300 · 2026-11-03
- Claude gets a Clio connector tapping 1B+ legal documents with citations — EJWalters · 2026-11-03
- DeepSeek open-sources dsh-superagent, an agent built to minimize human attention — daniel_mac8 · 2026-11-03
- UIBunny ships 8,000 real website screenshots from 350 products, MCP access for agents — coiboi48 · 2026-11-03