Live data from Hacker News

What OpenAI did when ChatGPT users lost touch with reality

nytimes.com

351–360 of 548 posts

Re: What OpenAI did when ChatGPT users lost touch with reality

#351
post #244

It surprises me how hyper focused people are on AI risk when we’ve grown numb to the millions of preventable deaths that happen every year. 8 million people to smoking. 4 million to obesity. 2.6 million to alcohol. 2.5 million to healthcare. 1.2 million to cars. Hell even coconuts kill 150 people per year. It is tragic that people have lost their mind or their life to AI, and it should be prevented. But those using t…

Agree that it's ridiculous to talk about banning AI because some people misuse it, but the word preventable is doing a lot of heavy lifting in that argument. Preventable how? Chopping down all the coconut trees? Re-establishing the prohibition? Deciding prayers > healthcare? Our society is deeply uncomfortable with the idea that death is inevitable. We've lost a lot of the rituals and traditions over the centuries th…

> Chopping down all the coconut trees? ... It probably isn't worth trying to prevent deaths from coconut trees

Would "not walking under coconut trees" count as prevention? Because that seems like a really simple and cheap solution that quite anyone can do. If you see a coconut tree, walk the other way.

Re: What OpenAI did when ChatGPT users lost touch with reality

#352
post #3

One of the more disturbing things I read this year was the my boyfriend is AI subreddit. I genuinely can't fathom what is going on there. Seems so wrong, yet no one there seems to care. I worry about the damage caused by these things on distressed people. What can be done?

There's a post there in response to another recent New York Times article: https://www.reddit.com/r/MyBoyfriendIsAI/comments/1oq5bgo/a_... . People have a lot to say about their own perspectives on dating an AI. Here's sampling of interesting quotes from there: > I'd see a therapist if I could afford to, but I can't—and, even if I could, I still wouldn't stop talking to my AI companion. > What about those of us who a…

I am thinking about the last entry. I'll be addressing them in this response.

If you're searching for emotional safety, you probably have some unmet needs.

Fortunately, there's one place where no one else has access - it's within you, within your thoughts. But you need to accept yourself first. Relying on a third party (even AI) will always have you unfulfilled.

Practically, this means journalling. I think it's better than AI, because it's 100% your thought rather than an echo of all society.

Re: What OpenAI did when ChatGPT users lost touch with reality

#353
Can’t we use LLMs as models to study delusional patterns? Like, try things that are morally questionable to try on a delusional patient. For instance, LLM could come up with a personalized argument that would convince someone to take their antipsychotics, that’s what I’m talking about. Human caretakers get frustrated and burned out too quickly to succeed

Re: What OpenAI did when ChatGPT users lost touch with reality

#354
The headline reads like a therapy session report. 'What did they do?' Presumably: made more money. In seriousness, this is the AI industry's favorite genre—earnest handwringing about 'responsible AI' while shipping products optimized for engagement and hallucination. The real question is why users ever had 'touch with reality' when we shipped a system explicitly trained to sound confident regardless of certainty. That's not lost touch; that's working as designed.

Re: What OpenAI did when ChatGPT users lost touch with reality

#355
This is exactly how natural language is meant to function, and the intervention response by OpenAI is not right IMO.

If some people have a behavior language based on fortune telling, or animal gods, or supernatural powers, picked up from past writing of people who shared their views, then I think it’s fine for the chatbot to encourage them down that route.

To intervene with ‘science’ or ‘safety’ is nannying, intellectual arrogance. Situations sometimes benefit from irrational approaches (think gradient descent with random jumps to improve optimization performance).

Maybe provide some customer education on what these systems are really doing, and kill the team that puts in response, value judgements about your prompts to give it the illusion you are engaging someone with opinions and goals.

Re: What OpenAI did when ChatGPT users lost touch with reality

#356
post #328

Earlier quoted context omitted.

It will probably increase the number of people deemed useless by the economy and the death rate of those people will be high. 1% of the world is over 800m people. You don't know if the net impact will be an improvement.

1% would be 80m not 800m. Still a lot of people but not 1/8 of the world population.

Sure, I stand corrected.

Re: What OpenAI did when ChatGPT users lost touch with reality

#357

It surprises me how hyper focused people are on AI risk when we’ve grown numb to the millions of preventable deaths that happen every year. 8 million people to smoking. 4 million to obesity. 2.6 million to alcohol. 2.5 million to healthcare. 1.2 million to cars. Hell even coconuts kill 150 people per year. It is tragic that people have lost their mind or their life to AI, and it should be prevented. But those using t…

People see that the danger will grow exponentially. Trying to fix the problems of obesity and cars now that they're deeply rooted global issues and have been for decades is hard. AI is still new. We can limit the damage before it's too late.

> We can limit the damage before it's too late.

Maybe we should begin by waiting to see the scale of said so-called damage. Right now, there have maybe been a few incidents, but there are no real rates on "oh x people kill themselves a year from ai" and as long as x is still that, an unknown variable, it would be foolish to speed through limiting everybody for what can be just a few people.

Re: What OpenAI did when ChatGPT users lost touch with reality

#358

It surprises me how hyper focused people are on AI risk when we’ve grown numb to the millions of preventable deaths that happen every year. 8 million people to smoking. 4 million to obesity. 2.6 million to alcohol. 2.5 million to healthcare. 1.2 million to cars. Hell even coconuts kill 150 people per year. It is tragic that people have lost their mind or their life to AI, and it should be prevented. But those using t…

Agreed - Really surprising this article didn't cover the flip side - how many lives have been saved due to having an instant source of truth in your pocket.

Source of "truth".

Re: What OpenAI did when ChatGPT users lost touch with reality

#359
post #56

Earlier quoted context omitted.

Claude has a sycophancy problem too. I actually ended up canceling my subscription because I got sick of being "absolutely right" about everything.

I've had fun putting "always say X instead of 'You're absolutely right'" in my llm instructions file, it seems to listen most of the time. For a while I made it 'You're absolutely goddamn right' which was slightly more palatable for some reason.

Have it say 'you're absolutely fucked'! That would be very effective as a little reminder to be startled, stop, and think about what's being suggested.

Re: What OpenAI did when ChatGPT users lost touch with reality

#360

This is exactly how natural language is meant to function, and the intervention response by OpenAI is not right IMO. If some people have a behavior language based on fortune telling, or animal gods, or supernatural powers, picked up from past writing of people who shared their views, then I think it’s fine for the chatbot to encourage them down that route. To intervene with ‘science’ or ‘safety’ is nannying, intellec…

I think it’s a silly take. Companies want to avoid getting bad PR. People having schizophrenic episodes with ChatGPT is bad PR.

There are plenty of legitimate purposes for weird psychological explorations, but there are also a lot of risks. There are people giving their AI names and considering them their spouse.

If you want completely unfiltered language models there are plenty of open source providers you can use.

Post reply on HN