3 ms·

Sounds like an education problem ? This kind of spam is an annoyance, but you learn to filter it out soon enough. Stack Overflow is a great example : its valu

Sounds like an education problem ? This kind of spam is an annoyance, but you learn to filter it out soon enough. Stack Overflow is a great example : its value is not so much in giving you a quick answer, but even more because you will see several answers, and the discussion around them. And yes, sure, convincing malicious actors there are always possible, but not probable. Same thing with Wikipedia's Talk pages, blog's comments (with the censorship issue), or third party comment sites like HN (with the platform issues). I'm much more concerned about artists that make works for their aesthetic value losing their already precarious livelihood to AIs. And even considering that it's sometimes feels like it's worth it : https://youtube.com/@D4BadRadio https://youtube.com/@D4BadRadio (who would have paid enough for this to be made by artists ? nobody)

palata 3y ago
> This kind of spam is an annoyance The problem is that LLMs are a whole new level of spam, orders of magnitude more powerful than anything we know. Before LLMs, you could not generate millions of comments for e.g. TripAdvisor without the users realizing that those were generated. With LLMs, users cannot say anymore if a review was written by a human or not. You just cannot trust any anonymous review you find online anymore. If Google cannot automatically discriminate between spam and genuine content, then Google is broken. And one fundamental promise of LLMs is that it is (or will be) powerful enough to reach that point, and therefore to break Google.
- BlueTemplar 3y ago
  I don't think that your Trip Advisor example is a good one, because there's already a threshold there : having to make a Trip Advisor account (and keep it from being flagged). Unless TripAdvisor's reviews are actually anonymous rather than pseudonymous ?! Isn't that the kind of super-targeted, high cost / high reward scam that smarter bots don't help with because its automation doesn't help much ? Also, if even pseudonymous accounts can be (meaningfully) penetrated by bots, then what prevents article's deep web from being penetrated and destroyed too ? (In fact we've already seen it happen on the likes of Facebook years ago, its faster, pseudo-ephemeral nature making things worse I guess ?)

palata 3y ago

> This kind of spam is an annoyance The problem is that LLMs are a whole new level of spam, orders of magnitude more powerful than anything we know. Before LLMs, you could not generate millions of comments for e.g. TripAdvisor without the users realizing that those were generated. With LLMs, users cannot say anymore if a review was written by a human or not. You just cannot trust any anonymous review you find online anymore. If Google cannot automatically discriminate between spam and genuine content, then Google is broken. And one fundamental promise of LLMs is that it is (or will be) powerful enough to reach that point, and therefore to break Google.

BlueTemplar 3y ago
I don't think that your Trip Advisor example is a good one, because there's already a threshold there : having to make a Trip Advisor account (and keep it from being flagged). Unless TripAdvisor's reviews are actually anonymous rather than pseudonymous ?! Isn't that the kind of super-targeted, high cost / high reward scam that smarter bots don't help with because its automation doesn't help much ? Also, if even pseudonymous accounts can be (meaningfully) penetrated by bots, then what prevents article's deep web from being penetrated and destroyed too ? (In fact we've already seen it happen on the likes of Facebook years ago, its faster, pseudo-ephemeral nature making things worse I guess ?)