Hacker News Reader

Comment by delichon | Hacker News Reader

There is an epistemic silver lining. This is in fact a Red Queen's race that cannot be won. So in the end the only solution is to evaluate the text on its own merits without reference to the writer's status, because that status can no longer be reliably detected. For a public feed like this one, the only alternative is to ignore it. The fire hose of data will inevitably become ever more fecal. We can only walk away from it or be more careful about the pearls we pluck out. It ends well only if we get better at pearl detection.

View on HN

One way that I could imagine a human-only HN could evolve in the coming AI wasteland: motivated individuals join small local groups and are validated face-to-face at meet-ups. Local trusted leads gatekeep their chapter’s posts, and this scalable moderation works up the tree. Bad leaves get culled out reasonably fast, maybe there’s some controls at the top level that let you see more content “lower down the tree” if you’re ok with lower SNR. Latency to get a post widely distributed grows but I don’t see that as a massive problem.

You're giving me flashbacks to PGP key signing parties.

I do like your idea, though.

Bring back the key signing parties!

p.s. @patrickmay: jinx!

In my recent experience, local meetups and groups are unexpectedly more prone to self promotion and low effort spamming.

Local groups have a problem where members admit their friends or pressure others into inviting their friends who are not a net positive, but it feels too impolite to refuse or to kick someone out. Meeting someone in person also develops a sense of a social bond that makes it harder to downvote or flag their posts.

Local groups have always been a haven for affinity fraud, too. Running a scam is easier when you can smile, be charismatic, and pretend to be a personal friend before springing your ask on to your victims.

This sounds like failure of leadership. Our coding meetups are already implementing what the GP suggested [0] and we also enforce our written guidelines (in this case, politely removing the bad eggs.)

[0] https://handmadecities.com/memos/HMC-Memo-004-Meetup-Hosts.p...

You've described a conference as well, for what that's worth.

> coming AI wasteland: motivated individuals join small local groups and are validated face-to-face at meet-ups. Local trusted leads gatekeep their chapter’s posts, and this scalable moderation works up the tree. Bad leaves get culled out reasonably fast,

Wow this is really cyberpunk.

I'll bring my Yubikey!

I've been thinking the same. One way to moderate is to bring back physical consequences.

I'd also like to see an "Order of the White Lotus" community (or Fight Club if you prefer) where people who collectively agree to not use AI against each other can come together. They can still use AI (i.e. out of necessity) just not with other members knowingly.

I suspect whatever form it takes the stakes will be very high to hack yourself into and pollute the space. So the more successful the community becomes, the harder it is to keep in order.

> So in the end the only solution is to evaluate the text on its own merits

This falls apart as soon as you realize that evaluating the text requires far more effort than generating it. If you're spending 2 minutes reading text that took 2 seconds to generate, you already lost.

That just means that you can only evaluate a smaller fraction of the data. If your goal is to do more than sample it, you've already lost.

"cannot be won" "only solution" "only alternative". sorry, no, that's too black and white. There are other solutions, even if they will only work for a couple of days/months/years.

Don't tell anyone, but I am secretly in charge and open to suggestions. Spill.

We can relentlessly bully anyone using phrases like "Red Queen's race" unironically. Measly human resistance against the vapid strip-miners of semantic value.

You mean that you don't believe that we are in co-evolution with AI? Because otherwise it is a Red Queen's race, and it is a useful frame for understanding. For example we can make it a race between symbiotes.

If you are Sisyphus, the fact that the hill is infinite is useful when planning your day.

I don't believe you are competent enough to be making those assessments.

Agreed. Merit is the only fair solution. If OP noticed a garbage post, that means they evaluated a post on merit and decided it was garbage. So it works.

We have genAI generating videos and the quality sucks compared to human produced and filmed content. People call it out and nobody is going to watch a genAI movie at the theater or binge a genAI TV show. Merit based filtering.

GenAI for music is not as good as human-generated music either. Not a single AI song from Suno or Udio has reached the top40. Not even one. 100% of the songs are human because they are evaluated on merit.

We have SWE and agentic benchmarks to evaluate coding LLMs on merit.

Disclaimer: I am a new account.

> Disclaimer: I am a new account.

Welcome. Illegitimi non carborundum.

The thing is, I can read something that's really terribly written and still extract useful information from it. (Suppose, for example, an LLM was directed to synthesize information from some sources that I wouldn't have thought of doing; or a submission simply makes me aware of a blind spot I had. Or I look up documentation and find something that's incredibly verbose and full of marketing-speak, but the code samples look reasonable and can be verified by testing and/or cross-reference.)

This comment uses a lot of big words but it’s full of fallacies.

The HN user base is not perfect at detecting LLM content but a lot of it does get flagged and downvoted eventually. About once a day I’ll click on a link, realize it’s AI slop, and go back to HN to flag it but discover that it’s already flagged.

If you turn on showdead you can see all of the comments from LLM bots that have been discovered and shadowbanned.

The fallacy in the comment above is simple: It’s taking the current situation and extrapolating to an extreme future, then applying the extrapolated future prediction on to the current situation. The current situation does not represent the extreme future predicted. A lot of the LLM content is easily spotted and a lot of it is a waste of time to read, therefore it’s right to police and ban it. Even if imperfect.

Earlier today I found something that impressed me as awful slop, but I was hesitant to flag the submission because as far as I could tell it got the facts right (I didn't try to verify some details of who was involved with what, but I was familiar with the proposals the article was discussing).

I'm somewhat keen to adopt ATProto's feed generators and/or labeller concepts to create an alternative /new and comment prioritizer

> The fire hose of data will inevitably become ever more fecal. We can only walk away from it or be more careful about the pearls we pluck out. It ends well only if we get better at pearl detection.

I'm not sure we can. Imagine an AI that 1) creates multiple accounts, 2) spews huge numbers of comments, 3) has accounts cross-upvote, and then 4) gets enough karma on multiple accounts to get downvote privileges. That AI now controls the conversation. Anything it doesn't like, it can downvote to death.

I mean, I'm sure that HN has a "voting ring" detector, but an AI could do this on a sufficient scale to be too large to register as one cohesive group. And I think HN has a "downvote brigading" detector, but if the AI had enough different accounts, I'm not sure that would trigger, either.

The best chance to detect it is just on volume (or perhaps on too many accounts coming from the same IP address or block). But if the AI was patient, I'm not sure even that would work.

That's depressing. I don't want HN to become a bot playground, with humans crowded out. But I'm not sure we can stop it, if it was done on a large enough scale.

Reply on news.ycombinator.com