Comment by Bender
1 day ago
The vast majority of programs submitted to HN are rip-offs of existing programs that are being re-written by LLM's and not even changing the name in most cases. Each day I am flagging more submissions than I used to in a month or two. People are absolutely crapping up this site with LLM plagiarized ideas and rewrites of existing code converted to other languages instead of doing pull requests of existing programs and enhancing them assuming LLM's can do such a thing. The voting ring detector probably needs some tuning as well. I can't tell if the goal is to poison the search engines and AI platforms so nobody can find the original open source programs or what else may be going on. Robert Hanlon said, never attribute to malice..., well I do.
It's a transitional situation and will be that way for quite a while. We're doing what we can. For example, the measures described here have stemmed the runaway growth in low-quality Show HNs:
https://news.ycombinator.com/item?id=47300772 (March 2026)
There's still a quality problem, but (a) that's always the case, (b) at least we aren't drowning, and (c) the community needs growth, just not runaway growth.
Major new tools like LLMs are inevitably going to get widely used, as they should. Figuring out what the best uses are will take time. Figuring out how to share what one is doing with them is an unsolved problem.
I see others also mentioned the voting ring detector. I think they're correct that something needs adjustment, but I don't know how it works exactly so it's hard to suggest thresholds. The patterns I am observing is that the LLM generated clones will get instantly upvoted multiple times I assume by other warmed up LLM accounts. Back in the days of running phpBB I would mitigate some of that with ranks and I assume there must be some parallels to that with account age and comments. Perhaps some math to lower the weight of upvotes by newer accounts, or the newer account being the upvote target so they can still gain some traction but slower. Perhaps one of the factors could be if the submission is for one of the public git repo sites AND it's a newer account it could have a divisor in the voting formula if it is not already.
The risk I see is that if the goal is growth but the low quality submissions are drowning out other submissions that could negate growth as some of the newer accounts that legit try to be part of a community would just drift away as they are in the poisoned well. I too struggle to think of a way to separate them out of the noise without creating a system that would just be gamed by the LLM's. On one hand if the system requires the regulars to "vouch" so to speak it will create little elitist bubbles whereas too much tweaking to algorithms will just be detected and gamed by LLM's.
Out of curiosity, are the LLM posts coming from residential and mobile addresses or from AI data-centers themselves? If it's not already that could be yet another weighting factor. And/or AI user-agents as a weight. There are many bot signals that could be weighting or division factors. Bots are easy to spot from the server.
The LLM posts that I'm looking at are definitely coming from normal-user IP addresses. There are exceptions, but the rate of those doesn't seem higher than usual. Outright bot/agent posting, as far as we can tell (and we may be wrong!) seems to have down-ticked since the flurry earlier this year (openclaw and so on).
My gut feeling is that this issue isn't much affected by voting rings, which is too bad, because we have a lot of experience with those. If all that was needed here was another round of work on the ring detector, I would be less worried.
It's a moving and blurry picture, but judging by the users that tomhow and I interact with—which is a lot of users! though still only a small sample—the overwhelming majority of these posts are coming from real people with good intentions, who have no idea of the mismatch between what they're posting and the culture of the community.
2 replies →