How this list was made
Every Hacker News comment from January 1 to November 12, 2026 was scanned for links. A link counts once per person, so one enthusiastic commenter can't push it up. Links to the thread's own article don't count: that's discussion, not a recommendation. Social media, news sites, shops and issue trackers were filtered out.
| HN comments, Jan 1 – Nov 12 2026 | 3,301,635 |
| comments containing a link | 317,158 |
| links in those comments | 415,391 |
| URLs linked by 3+ different people | 8,640 |
| mentions labeled by the model | 46,890 |
| resources listed here | 647 |
Which links are recommendations?
Most links in comments aren't recommendations: they're evidence, sources, jokes or things people disagree with. An AI model (OpenAI's gpt-6-luna decision model) read every comment that mentions one of the links and answered fixed questions: is the commenter recommending it? Is the page something a software engineer can learn from? What topic, format and level is it? It returns a probability for each answer.
A resource's score is the expected number of different people who recommended it: the sum, over commenters, of the probability that they were recommending it, times the probability that the page is learning material. Resources with a score of at least 2 that are more likely than not learning material are listed.
How accurate is it?
I labeled 79 mentions by hand and tuned the question wording on part of them. On 33 mentions the wording was never tuned on, the model was right about "is this a learning resource" 88% of the time and "is this a recommendation" 88% of the time. Topic and level labels are less reliable. Expect some mistakes, and use the HN links on each resource page to read what people actually said.
Visitor votes
Votes from visitors are shown separately and never change the HN ranking. Voting needs a one-time, usually invisible, human check (Cloudflare Turnstile). I store a random voter id per browser, and a salted hash of your network address for rate limiting that is discarded after a day. No accounts, no tracking cookies.
Who made this
I'm Mateusz Gazdziak. I built this because some of the best answers to "what should I learn next?" are buried in Hacker News comment threads, and I wanted them in one place.