3 ms·

This problem isn't going to be unique to Twitter even if we're seeing it there first. Any site with user generated content will be chasing a never ending battle

This problem isn't going to be unique to Twitter even if we're seeing it there first. Any site with user generated content will be chasing a never ending battle against bots churning out LLM content. There's really no way around it from what I can tell. How does one actually recognize and block not content online in an automated way? When the platform is digital and we now have digital means of generating content, the battle is lost IMO. Maybe it could be delayed by some kind of platform or regulatory change requiring state issued ID for all internet users. I think that has come up before and it feels very similar to what a few Googlers proposed (before the community dog piled on it). That's not a version of the internet I'd want to use though, and if that's the only avenue for avoiding a web full of bots it sure seems like the days are numbered.

rsynnott 3y ago
In general you'd deal with it by blocking where possible, and where not possible, prioritising comment threads by engagement. Twitter's special sauce for the bots is its pay-for-attention model; note that all the bots in the article are blueticks, and thus get promoted to the top of the replies. This makes it a far softer target than a normal social network, where uninteresting bot comments would generally end up deprioritised due to lack of interest.
numpad0 3y ago
> How does one actually recognize and block not content online in an automated way? By gauging content popularity. People are good at consuming and obsoleting something abstract as an entire generative pipeline as content of its own. GenAIs and LLM had been doing great among its users but not reaching outside of it, and as required upkeep costs to fight its obsolescence grows non-linearly I'd predict it'll run out of investment sooner or later(before factoring in negative PR). Hard ID systems don't work. If it worked Facebook wouldn't have acquired Instagram and wouldn't be in competition against Twitter in the first place.
- _heimdall 3y ago
  > By gauging content popularity Isn't that already roughly how Twitter's feed algorithm works? I don't use Twitter, but from both the article and comments here it sounds like many people feel their feed is already full of LLM content.
AnimalMuppet 3y ago
How does HN deal with it? HN has user-generated content, after all, and it isn't (yet) overrun with bots. On HN, users downvote and flag junk content. dang shadowbans junk accounts. (I don't know how much of dang's bandwidth this takes up, but it could become a problem. dang does not scale.)
- _heimdall 3y ago
  HN doesn't really have an incentive model for not content. Personally I don't see down votes or flags as a not protection, they're meant to help patrol against low quality content or content that breaks moderation rules. Overloading it for bots may work for a while, but only for as long as not content is recognizable and/or bad quality. How would we know to flag not content when it looks just like a real person may have written it? And how do we know quality content isn't already being written by bots?
layer8 3y ago
> Any site with user generated content will be chasing a never ending battle against bots churning out LLM content. I have a hard time imagining this happening on HN.
- _heimdall 3y ago
  Is there any technical reason for it being hard to imagine, or just the lack of incentive for posting LLM content here?
  - layer8 3y ago
    Lack of incentive plus likelihood of being downvoted or swiftly banned by dang. Or for submissions, unlikelihood of being upvoted. Of course, if LLM content is so good that it survives on HN, then it may actually be welcome.

rsynnott 3y ago

In general you'd deal with it by blocking where possible, and where not possible, prioritising comment threads by engagement. Twitter's special sauce for the bots is its pay-for-attention model; note that all the bots in the article are blueticks, and thus get promoted to the top of the replies. This makes it a far softer target than a normal social network, where uninteresting bot comments would generally end up deprioritised due to lack of interest.

numpad0 3y ago

> How does one actually recognize and block not content online in an automated way? By gauging content popularity. People are good at consuming and obsoleting something abstract as an entire generative pipeline as content of its own. GenAIs and LLM had been doing great among its users but not reaching outside of it, and as required upkeep costs to fight its obsolescence grows non-linearly I'd predict it'll run out of investment sooner or later(before factoring in negative PR). Hard ID systems don't work. If it worked Facebook wouldn't have acquired Instagram and wouldn't be in competition against Twitter in the first place.

_heimdall 3y ago
> By gauging content popularity Isn't that already roughly how Twitter's feed algorithm works? I don't use Twitter, but from both the article and comments here it sounds like many people feel their feed is already full of LLM content.

AnimalMuppet 3y ago

How does HN deal with it? HN has user-generated content, after all, and it isn't (yet) overrun with bots. On HN, users downvote and flag junk content. dang shadowbans junk accounts. (I don't know how much of dang's bandwidth this takes up, but it could become a problem. dang does not scale.)

_heimdall 3y ago
HN doesn't really have an incentive model for not content. Personally I don't see down votes or flags as a not protection, they're meant to help patrol against low quality content or content that breaks moderation rules. Overloading it for bots may work for a while, but only for as long as not content is recognizable and/or bad quality. How would we know to flag not content when it looks just like a real person may have written it? And how do we know quality content isn't already being written by bots?

layer8 3y ago

> Any site with user generated content will be chasing a never ending battle against bots churning out LLM content. I have a hard time imagining this happening on HN.

_heimdall 3y ago
Is there any technical reason for it being hard to imagine, or just the lack of incentive for posting LLM content here?
- layer8 3y ago
  Lack of incentive plus likelihood of being downvoted or swiftly banned by dang. Or for submissions, unlikelihood of being upvoted. Of course, if LLM content is so good that it survives on HN, then it may actually be welcome.