[bilbo0s]

This is kind of a big problem.

Because on the one hand, sycophancy is not really what you want to do for people in mental and emotional crisis. On the other hand, not being sycophantic is not really what you want to do for people in mental and emotional crisis.

There are professionals who speak to people in crisis for a reason. That's because it's fraught with pitfalls and trapdoors that take the situation from "mental and emotional crisis" to "tactical emergency" in a heartbeat.

I know that no one wants to hear this, but ChatGPT should probably be listening for people in crisis and, well, maybe not calling the cops, but maybe if there is a crisis line in their jurisdiction? A suicide hotline or something?

I don't know? But having an LLM out trying to handle that on its own just seems like a very bad idea.

[thewebguyd]

> I know that no one wants to hear this, but ChatGPT should probably be listening for people in crisis and, well, maybe not calling the cops, but maybe if there is a crisis line in their jurisdiction? A suicide hotline or something?

Doesn't necessarily even need to call (particular in case of false positives) but there absolutely should be detection and a cutoff switch, where the chatbots just refuse to continue the conversation and then print out the hotline numbers (much like with reddit cares messages).

I'm generally not in favor of censorship or overly protective safeguards on LLMs, but maybe it's needed for hosted models/services that are available to the masses.

But before they get locked down more, we should try some legislation to limit how they can be marketed and sold. Stop letting OpenAI, etc. call the models "intelligent" for one. Make the disclaimers larger, not just small print in the chat window but an obvious modal that requires user agreement to dismiss - disclaim that it's a predictive engine, it is not intelligent, it WILL make mistakes, do not trust its output. Make it clear during the chat session over and over again, and then have a killswitch for certain paths.

[sho_hn]

I think a good first step would be if a ChatGPT user account could easily enter emergency contact information the system would notify in scenarios like this, making the escalation opt-in.

The moderation tech is already there, and if there's even a small amount of mentally ill who would fill this in on a good day and be saved by it on a bad day / during an episode, it'd be worth it.

[cm2012]

This may reduce liability but if your answer to someone in crisis is "dont talk to me, call the official number", that will be considered a very negative user experience.

[idle_zealot]

I bet you could train a supervisor classifier to run on chats and once it reaches a certainty threshold that the users is spiraling it could intervene and have the interface respond with a canned message directing the user to a help line. Of course, OpenAI wouldn't do this, because that would involve admitting that its bot can exacerbate mental health issues, and that LLM therapy is more harmful than helpful, which cuts against their AGI/replace human workers sales pitch. Their product isn't the LLM, it's trust in the LLM.

[seunosewa]

You can't force people to seek help. Best to help them as much as you can while trying to persuade them to seek help.

[idle_zealot]

The idea is that the canned message would be an attempt at persuading them. I really don't trust an LLM prompted to persuade someone to seek therapy would yield better results.

[seunosewa]

Modern thinking models can be trusted to follow nuanced safety instructions. Models like ChatGPT-4o can't. They will make bizarrely inaccurate statements from time to time.

[cm2012]

My wife has psychosis and Gpt5 consistently is a voice of reason and grounds her. She shows me her GPT chats to get me caught up and its so nice to have something that can patiently listen to her when she repeats herself a lot.

In the meantime Ive had two therapists that we ended with since they didnt help the condition, and we're very expensive.

[bilbo0s]

And that's great! It should be celebrated.

But we shouldn't set potential school shooter intervention policy based on the experience of a single person in crisis with GPT5. We have to set it on the basis of people who may be in crisis and may not have the support network, of, say.. a husband for instance.

Now we also shouldn't set it based on the worst case. But at the mean it's clear many people don't have the supports that your anecdata point presupposes. And at the same time we should try to find answers there that aren't simply, "Hey ChatGPT, report this person to the cops!" (Or maybe that is the answer? I'm not an expert, so I don't know? But it strikes me that we could all be trying some other things before we go all the way to the law enforcement backstop.)

[ACCount37]

I think "not being sycophantic" is the lesser error in just about every scenario. You do more harm by reinforcing delusions than you could ever do with a dismissive attitude of "no, that sounds psychotic, you should get your head checked".

But a big part of the issue is that OpenAI wants user engagement - and "not being sycophantic" goes against that.

They knew feeding raw user feedback into the training process invites disaster. They knew damn well that it encourages sycophancy - even if they somehow didn't before the GPT-4o debacle, they sure knew afterwards. They even knew their initial GPT-5 mitigations were imperfect and in part just made the residual sycophancy more selective and subtle. They still caved to the pressure of "users don't like our update" and unrolled a lot of those mitigations.

[sho_hn]

"not being sycophantic" should also be an easy decision considering that OpenAI isn't ad-funded and doesn't need to directly optimize for session length/engagement.

[kiba]

They have no underlying strategy or vision on how they mitigate harms and improve trust in their product, so of course they roll back on anyone that screams bloody murder as opposed to addressing root causes.

[wrycoder]

Perhaps flag the account and switch out the AI instance for Boring-BobGPT and further monitoring?

[lazide]

Just what we need, the LLM ‘Reddit cares’. Notably, that is a mess too.

Also plenty of those hotlines are BS, or don’t work, or flat out don’t exist for given locales, etc.

The biggest issue is that LLM’s can act like a person, but aren’t a person, and fundamentally this causes problems. Especially for people that are already borderline or fully crazy.

[ACCount37]

The biggest issue is that LLMs can act like a person indeed. And the kind of person an LLM acts as? Well.

When you train on raw user feedback, you can easily end up wiring some incredibly undesirable patterns into your AI. Resulting in things like an AI that never wants to contradict its user, and always wants to support its user in everything, and always wants the user to like it. See GPT-4o for the kind of outcomes that results in.

[mathiaspoint]

Either people can manage themselves or they can't and the ones that can't probably need to be institutionalized for their and everyone else's safety.

[sho_hn]

In reality it's not that binary. Plenty of people can manage themselves, except during times when they can't, and then they need help.

It'd be a good start if services let you enter emergency contact info, making escalation opt-in.

[mathiaspoint]

That is actually absurd. If they're insane enough to think the world is conspiring against them they wouldn't surrender that way.

[sho_hn]

No, it's not, because, again, it's not as binary as you think it is. There's plenty of people who have good times and psychotic episodes, and during the latter they can be not collected and organized enough to dismantle safety networks they put in place during good times for exactly that reason.

A feature like that definitely wouldn't help everyone, but it might help some. If it was sold as a blanket solution that would be indeed absurd however.

[ayewo]

> On the other hand, not being sycophantic is not really what you want to do for people in mental and emotional crisis.

Having trouble parsing the double negation in your comment.

Sorry, I’ve had a long day :)