undefined | Better HN

0 pointslondons_explore1mo ago0 comments

Met a chap who wrote exactly like an AI all over reddit. Had 10k+ posts going back a decade, all in the same helpful style with em-dashes. Telephoned him to check he was a real person.

Turns out that certain really helpful reddit posters respond in exactly the same way AI companies wish their models would respond, and the RLHF process really reinforces their mannerisms despite being 0.00001% of the total training data.

I feel sorry for them - their recent post history is full of mods banning them for being a bot.

0 comments

westurner1mo ago

I have had this experience.

My posts are link dense because citations are worth sharing.

It's really frustrating because they're half-assedly alleging in ignorance when I'm taking the time to do something more like actual research and sharing.

I don't want to assume it's an inadequacy or an inferiority self-soothing response.

Perhaps more troubling than AI replacing actual critical thought is "that's AI so it's bunk" replacing actual dissent and argumentation.

Someday I'll compare occasional archives of my outbound communications to see what's been flagged and censored for example "because that's AI".

8cvor6j844qw_d61mo ago

Ouch. Yeah I imagine AI models get fed with proportionally more of their content, hence looks and reads like their style.

j / k navigate · click thread line to collapse