Hacker News Reader

Comment by lokar | Hacker News Reader

In that case was the shooter using the LLM to plan the attack?

That could be a meaningful difference

That's what the suit alleges. Also that it had been automatically flagged to an internal human review team who had concluded that there was a credible, specific risk of harm to others and recommended notifying police, and that it was OpenAI leadership that rejected that recommendation.

You can see it all in GP's link. It's very readable.

... Wow. Well that's not at all what I thought happened.

I figured it just didn't get flagged properly.

So mentioning killing in passing is fine, but planning with the tool is not? How should LLM providers draw the line? I'd prefer if they didn't scrutinize any of my inputs to see if I am committing a felony or not in the first place - but that ship has sailed.

...yes?

They draw the line using a team of humans who evaluate whether the threat is real and give a recommendation to management.

If you operate a gun store and someone reveals to you that they intend to use your product to shoot up a school you have a similar moral obligation, I think. This isn't a special case.

But it is, it somehow only applies to LLMs. What about doc apps? Todo lists? Chat apps?

There are tons of hosted app services where you can write private expressions. Why are only LLMs subject to monitoring?

You're really stretching here for some reason I can't fathom, but yes - if you write a credible threat to shoot up your school on a crumpled up paper and someone finds it, they should report it. I'm not sure that complicating this with which kind of app it's written in is all that useful ...

I don’t see how writing something you clearly intend to be private can be a threat.

Is this a point you're arguing seriously? If you find someone's plans to, for example, shoot up a school crumpled in a note on the ground that should be...ignored? Or what? I struggle to comprehend what you're actually arguing for here.

I think I pretty much agree with you that if someone finds evidence of danger to someone else's life then it is a fair expectation that they do something with that evidence. And despite that I still believe that is not a good enough reason to eliminate someone's right to privacy, and yet still that right doesn't mean that we should be able to demand privacy in all products. I think it is fairly clear that these big LLMs are not private products and we shouldn't expect them to be, but I'm still in support of offering private alternatives that people could use without having every input scrutinized.

I guess I just see a lot of "X, but with an LLM" getting different treatment than just X recently, and it's confusing.

I don't think that most (any?) of these other cloud services monitor for this stuff, or that government thinks they have an obligation to do so.

How many mass shooters left diaries, letters, memoirs containing threats? I don't see why you don't see.

Innocent until proven guilty. Every citizen has rights until there is reasonable suspicion, probable cause, etc that they are commiting a crime.

The right of the people to be secure in their persons, houses, papers, and effects, against unreasonable searches and seizures, shall not be violated...

The court is going to have a look at it.

If you want to be secure in your papers, maybe don't type them into a giant data collecting machine.

How many of them drank water and/or respired oxygen?

The false positive rate of the test that you suggest is frankly atrocious.

Uh, no, text you write in private should not be automatically turned over to the police.

Sure; but writing to an LLM is not "in private". That's the point of the original post.

The average consumer thinks it is.

There's a big, big difference between "finding by happenstance" and "automated monitoring and analysis of everything written with invisible alerting of authorities". It's a step towards a digital panopticon and there are more than a few objections people would have to living under a digital panopticon.

If it’s just a journal, how is it different then Google Docs?

Reply on news.ycombinator.com