To be honest, personally I had no idea it was generated until the author said so at the end.

Makes me worry about my own reading comprehension, but I think what happened was that since it was posted on HN and got upvoted a lot, I simply assumed that anything that I didn't understand was not the writer's fault, but mine.

For instance, it was unclear from the post what the bitcoinforum experiment was about, but I just dismissed it as me not being attentive enough while reading.

At one point GPT-3 writes: "The forum also has many people I don’t like. I expect them to be disproportionately excited by the possibility of having a new poster that appears to be intelligent and relevant." Why would people he doesn't like be paricularly excited about a new intelligent poster? Again I just assumed that I missed the author's point, not that it was nonsensical.

Twice it refers to tables or screenshots that are not included, but it seemed like an innocent mistake. "When I post to the forum as myself, people frequently mention that they think I must be a bot to be able to post so quickly" seemed like another simple mistake, meaning to say that when he posted as GPT-3, people thought he was being too quick.

This is like a written Rorschach test, when I'm convinced that what I'm reading must make sense, then I'll guess at the author's intent and force it to make sense, forgiving a lot of mistakes or inconsistencies.

The second one doesn't sound like a mistake to me. Someone being able to consistently post so quickly is actually a valid sign of being a bot.

Really interesting. Looking back, I did the exact same thing at many of the same points.

It will be very annoying for e.g. forum moderators to determine whether first user posts are just a bit incoherent, or generated spam garbage.

That used to be a pretty annoying thing back in the days of IRC as a kind of DOS: run a bunch of bots that just replay a conversation from another channel. Engaging them fails, but is that because they are bots, or because they're just ignoring you?

The new kind can be more targeted for specific purposes. They could be excellent tools for trolling a forum, inciting flame wars and such.

That would require some more advanced tech though. I don't think GPT-3 can target divisiveness yet, especially since it would heavily depend on the community you're writing for, e.g. driving a wedge into the general population is very different than driving a wedge niches. The Linux vs Windows debate might get you engagement in a tech forum, but it'll fall short with social housing activists, and whatever issues they split on will probably not get you anywhere with the tech crowd.

I don't think it needs to understand what a divisive issue is to have an effect. If you've got a human operator who can pick a divisive enough prompt, this can dramatically increase their inflamations-per-hour because they don't need to compose the body text.

It's true that distinguishing these articles from ordinary jointed ramblings of poor writers would be hard. But I'm not sure what the benefit of filling forums with babble has to those running these models.

Bots offering idiocy and idiocy generally has done lots of damage. But by idiocy here I would quite carefully calculated cleverly polarized positions and I don't think just bot-rot would be enough (to maybe coin a phrase).