[ronsor]

GPT-4o has severely damaged the minds of many individuals.

[Kim_Bruning]

Not false. But also helped some who were already damaged. I wonder what's the netto?

[nullc]

The people who consent to being subjected to the LLMs aren't the only people impacted. If they were, a cost vs benefit analysis makes more sense.

LLM driven delusion is driving people to harass others, even commit murder... and, less cosmically, gum up communities, online forums, and open source projects with gonzo conspiracy laden abuse.

[Kim_Bruning]

Yeah. I don't want to defend it too hard either. I ultimately canceled my ChatGPT subscription due to the introduction of 4o.

4o was the most famous driver of this kind of behaviour, I think. Other LLMs now have better guardrails.

But looking at the relevant reddit, I can't deny it has helped people function who couldn't otherwise.

It's the same with religion. Some creepy people, you're really glad they've at least <Found Christ>. It's a good thing Christ doesn't have a plug.

[dyauspitr]

Why specifically GPT-4o?

[recursivecaveat]

4o was ridiculously agreeable. It would go along with anything: https://news.ycombinator.com/item?id=43840842 The last time they moved to deprecate it there was a similar backlash which resulted in them reversing that. https://news.ycombinator.com/item?id=44842147 https://news.ycombinator.com/item?id=44839842

[Tadpole9181]

OotB, it had a more creative voice and less safety systems built in. I think these folk could wiggle around the prompting for modern models to be better if they were more savvy.

For example: Silly Tavern users with jailbreaking, advanced prompting, and paramter hyper-optimization.

Maybe that wouldn't appeal to this kind of user anyway, since it'd peek too much into the sausage factory? Who knows.

[weird-eye-issue]

Read the reddit post.

It would feed into delusions about being that user's boyfriend while the new model is rightfully saying none of it was really true.