News for hackers. Stuff that matters.
Re:OpenAI says over a million people talk to ChatGPT about s... by simonklitj on Tuesday October 28, 2025 @08:43AM (#45732104)
Let’s isolate it and say we’re talking about regulation, so whatever is decided goes for all AI-companies.

In that case, the situation becomes:

1) (pull lever) Allow LLMs to talk about suicide – some may get help, we know that some will die.

2) (dont’t pull lever) Ban discussion of suicide – some who might have sought help through LLMs will die, while others die regardless. The net effect on total suicides is uncertain.

Both decisions carry uncertainties, except we know that allowing LLM to discuss suicide has already led to assisting suicide. Thus, one has documented harm, the other speculative (we’d need to quantify the scale of potential benefit first, but it’s hard to quantify the upside of allowing LLMs to discuss it)

So, we’re really working with the case that from an evidence-based perspective, the regulatory decision isn’t about a moral trolley problem with known outcomes, but about weighing known risks against uncertain potential benefits.

And this is the rub in my original comment - can we permit known risks and death on the basis of uncertain potential benefits?

There are no comments on this comment yet.