Hacker News Reader

Comment by whateveracct | Hacker News Reader

It seems like it wouldn't be hard to train a model on this era of 4chan and create a simulacrum. Posts would happen at a certain frequency with all the users being simulated. I wonder if you could even allow user posts and have the model reply in a realistic way (bullying and all)?

View on HN

The remarkable thing is how effective bullying is at moderating idiots. Astroturfing state actors aside, the site maintains a high level of on topic civility.

There's something inherently wrong with ordering posts by vote count, both because a confident/deceptive idiot can spread his lies like a virus and because it's too easy to game.

Slashdot solved this ages ago with its moderation system, vote moderation has been flawed since inception, because people downvote when they disagree not based on the quality of the post.

I'm reasonably sure this is the first time I've ever seen a claim associating civility with 4chan.

I suppose you need to be deep in the trenches to find anyone still willing to have discourse there. But as another commenter said, some of the niche boards, like subreddits, maintain some pretty high quality conversation.

On a small scale I've seen this work, and maybe early 4chan was like this.

Current-day, though? Nah, idiots attract more idiots and most of the site sucks save for a few niche boards/topics.

Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should

> It seems like it wouldn't be hard to train a model on this era of 4chan and create a simulacrum.

Out of curiosity, are there any models trained on HN posts?

You're using one right now

The problem is that the model we’re using now is locked in the world’s most secure HSM and you can’t use several copies at once.

Shadowbanned riding like the wind.. Gives model-citizen a whole new ring..

It works for X

I don't have the link handy but I do remember reading about an LLM project trained on 4chan.

https://en.wikipedia.org/wiki/GPT4-Chan

Reply on news.ycombinator.com