Ask HN: Does HN do anything to prevent AI-grade bots? | Hacker News Reader

Ask HN: Does HN do anything to prevent AI-grade bots?

5 pointsekianjo9 comments

Hi, since conversation between human users at HN is what makes the platform attractive at it is, I am wondering what HN is doing, or is going to do, when more sophisticated bots could land to parrot certain narratives or influence the conversation on important topics. Is there any way to detect such behavior?

View on HN

One of my favorite aspects of HN is that we judge people by what they say, not who says it. The minimalist profiles, without any profile pics, are a feature.

If a bot dispenses wisdom, then why not? It's still the responsibility of readers to check for truth and flag astroturfing and other violation of the guidelines.

Technically you do not have enough time in the world to check for the truth behind every story. That is precisely why propaganda works: share of voice overwhelming the possibility of dissent.

I don't know that it would matter. Propaganda, astroturfing and shilling campaigns are old news, and were old long before computer networks. I don't think there's a practical way of preventing such things.

For detection, fingerprinting would probably work for individual training sets or implementations, but with an advanced adversary (especially if they're politically rather than profit motivated), I'm not confident you could even whack-a-mole well.

If such a hypothetical bot really can pass as human what would you suggest the defense be? People are quite capable of pushing agendas themselves.

It has always been possible to outsource this work for a few cents. Whether it's a bot or cheap labor from an English-speaking third world country doesn't make that much of a difference.

I don't think HN is mainstream enough for people to push agendas. But on platforms like YouTube, etc this has been the norm for years.

GPT sucks at being like a human, it rapidly loses context

I can't help but to wonder why that is. Why couldn't chatbots and such (I guess bots that would make comments in some thread) just constantly remember everything that was said previously?

Trivial steps to reduce most of the spam / future spam:

1. Make the HN accounts scarce

2. And / or make them cost some bitcoin

For 2, a nation state or even a remotely big corp would have no problem footing the bill.

Reply on news.ycombinator.com