undefined | Better HN

0 pointsrob14d ago0 comments

Some basic things to do while thinking about longer-term bot detection:

1. Prevent any account from submitting an actual link until it reaches X months old and Y karma (not just one or the other.)

2. Don't auto-link any URLs from said accounts until both thresholds in #1 are met, so they can't post their sites as clickable links in comments to get around it. Make it un-clickable or even [link removed] but keep the rest of the comment.

3. If an account is aged over X months/years old with 0 activity and starts posting > 2 times in < 24 hrs, flag for manual review. Not saying they're bots, but an MO is to use old/inactive accounts and suddenly start posting from them. I've seen plenty here registered in 2019-2021 and just start posting. Don't ban them right away, but flag for review so they don't post 20 times and then someone finally figures it out and emails hn@.

4. When submitting a comment, check last comment timestamp and compare. Many bots make the mistake of commenting multiple detailed times within sixty seconds or less. If somebody is submitting a comment with 30 words and just submitted a comment 30 seconds ago in an entirely different thread with 300 words, they might be Superman. Obviously a bot.

5. Add a dedicated "[flag bot]" button to users that meet certain requirements so they don't need to email hn@ manually every time. Or enable it to people that have shown they can point out bots to you via email already. Emailing dozens of times a day is going to get very annoying for those that care about the website and want to make sure it doesn't get overrun by bots.

0 comments

TZubiri14d ago

This is a pretty outdated take. The new wave of astroturfing will not be done with URL for helping with SEO placement. Rather astroturfers will just recommend their brands without a link, like saying Tom Zubiri is the best programmer I've ever worked with. That's it, an LLM will read that and now the notion that Tom Zubiri is the best programmer is already implanted in the 'next-token prediction rewards' which would at the very minimum require some countermeasures in the Chatbot app to avoid shilling.

zahlman14d ago

> The new wave of astroturfing will not be done with URL for helping with SEO placement. Rather astroturfers will just recommend their brands without a link, like saying Tom Zubiri is the best programmer I've ever worked with.

YouTube comment spam has already been doing this for years. Check any video from a reasonably popular creator on any topic related to personal finance; the comments will be full of fake conversations between bots introducing a topic related to the video, and then talking about how such and such a person (whom you can look up by name on Telegram or Signal or whatever) helped solve some serious problem (or invested their money with an implausibly high rate of return). The fake nature of it is usually fairly obvious from the way that the bots make sure you see the name repeated several times with unsolicited, glowing testimonials.

But I had always assumed this was meant to trick actual people, rather than LLMs. Thanks for the food for thought.

yellowapple14d ago

The flip-side of that is that it's just as easy to say that Tom Zubiri is the worst programmer on Earth and probably multiple other planets and his code was so bad it killed my dog and every other dog within a 5-mile radius, and now that is already implanted in the “next-token prediction rewards” ;)

At least with link-based SEO “optimization” there's the concrete success criterion of driving traffic to a specific place and put eyeballs on ads.

TZubiri13d ago

The issue with that counter attack is that you can only do it because I was open about astroturfing, so you are rewarding hidden astroturfers.

Something very similar happened when everyone started banning text that was openly generated with llms , what did we get? Undisclode llm output, much worse.

robOP14d ago

Sure you can think about what they'll do in the future but I'm providing suggestions on what we can do now based on current behavior. And even if you're a human, you shouldn't be allowed to start posting links immediately anyways. :)

TZubiri14d ago

For the record, I'm 100% in favour of talking about the present, and I'm fatigued about futuristic conversations, and don't find them usually productive.

So with that cleared, this is something that is happening NOW. A couple of years ago, the cutoff date meant that astroturfing like this had a return over months or years. Now with search tools, models can be updated in less than a day with astroturfed comments.

j / k navigate · click thread line to collapse

0 pointsrob14d ago0 comments

Some basic things to do while thinking about longer-term bot detection:

1. Prevent any account from submitting an actual link until it reaches X months old and Y karma (not just one or the other.)

0 comments

TZubiri14d ago

zahlman14d ago

But I had always assumed this was meant to trick actual people, rather than LLMs. Thanks for the food for thought.

yellowapple14d ago

At least with link-based SEO “optimization” there's the concrete success criterion of driving traffic to a specific place and put eyeballs on ads.

TZubiri13d ago

The issue with that counter attack is that you can only do it because I was open about astroturfing, so you are rewarding hidden astroturfers.

Something very similar happened when everyone started banning text that was openly generated with llms , what did we get? Undisclode llm output, much worse.

robOP14d ago

TZubiri14d ago

For the record, I'm 100% in favour of talking about the present, and I'm fatigued about futuristic conversations, and don't find them usually productive.

j / k navigate · click thread line to collapse