mindbleach comments on ChatGPT Helped Tumbler Ridge Shooter Focus on Guns, Tactics, and Terror

[–] mindbleach@sh.itjust.works 3 points 1 month ago (4 children)

Write me the set of instructions that prevent all bad things from happening.

You can pretend we're using A Guy Instead, faking the bot part of the chatbot, so it's all real intelligent humans talking to one another. Show me what possible outline could stop someone from asking all dangerous questions, such that any crack in your effort makes you liable for what some crank does with that information.

Information which I will remind you is just scraped from the web. And hypermasculine military-fantasy novels. And dry encyclopedias of historical minutia.

permalink

fedilink

source

parent

hideshow 4 child comments

[–] nik282000@lemmy.ca 0 points 1 month ago (3 children)

Buddy, you are trying to argue that a business is NOT responsible for the text that was generated by it's service. If they included dangerous information in the training data (which was mostly acquired by violating copyright) then they are responsible for not filtering that training data. And here were safeties and they did flag the interaction, management overrode them and actively chose NOT to contact law enforcement. How much more liable do you want?

They haven't even gone public yet so I donno what horse you have in this race to make you side with a bunch of fucking tech bros. Their ideal future includes pulling the rug out from whatever industry you're in and leaving you to toil in whatever labour they can't automate.

permalink

fedilink

source

parent

hideshow 3 child comments

[–] mindbleach@sh.itjust.works 3 points 1 month ago (2 children)

Dangerous information like library books. Like the text of any Tom Clancy novel. These models, online or offline, just pick words based on what you write. You choose what comes out. If you ask for vile pornography, you don't get to blame the company for offending you, and if some prick asks for tactical advice 'in Minecraft,' the company did not proactively radicalize him. The chatbot can only repackage what's readily available to any pirate.

And here were safeties and they did flag the interaction, management overrode them and actively chose NOT to contact law enforcement.

That's a wildly different standard than 'no matter what they do any problem is entirely their fault.' You are trying to argue that even if management had told law enforcement, and this prick had a shootout with the arresting officers, the company would still be to blame. Their model still generated... what he fuckin' asked for. That's all it does.

All of these companies are fucked anyway. The bubble will burst. It'd be nice if the proud haters could conceptualize a milquetoast defense of the technology as anything but boot-slobbering devotion to specific high-profile cocaine enthusiasts. 'There's less work, oh no' is not a real problem. A child knows the answer. And if a local model lets that child slap together a program that surpasses what I could do, fresh from a college education in computer science - good. We've been trying to make that happen for sixty years.

permalink

fedilink

source

parent

hideshow 2 child comments

[–] nik282000@lemmy.ca -1 points 1 month ago (1 child)

Dangerous information like library books. Like the text of any Tom Clancy novel. These models, online or offline, just pick words based on what you write. You choose what comes out. If you ask for vile pornography, you don’t get to blame the company for offending you, and if some prick asks for tactical advice ‘in Minecraft,’ the company did not proactively radicalize him. The chatbot can only repackage what’s readily available to any pirate.

The majority of that text is being used in violation of copyright, but that's beside the point.

My point is that if I were to have see kid online say they were gonna shoot up a school, and then I spent the next week coaching them how to do so, I would be responsible for my involvement with the eventual shooting. Operating a service which does the same thing should entail the same responsibilities. OpenAI's failure to control their service does not absolve them from responsibility, if anything it shows how knowingly irresponsible they are.

They can easily implement stronger limits on what their models are allowed to say to a user BUT that would makes their service feel more like a computer and less like a person. A simple regex and blacklist of words and phrases could restrict topics. A second agent that only looks at the model output could be used as a nanny. Are you saying that it is reasonable that a teenager was able to bypass all the safeties of a billion dollar business?

And if a local model lets that child slap together a program that surpasses what I could do, fresh from a college education in computer science - good. We’ve been trying to make that happen for sixty years.

That says more about you than the LLM. LLMs can be used as an elaborate autocomplete but fully LLM written projects are known as slop for a reason, each small part may be complete and correct but as a whole it is just an unmaintainable mess of security issues and technical debt.

permalink

fedilink

source

parent

hideshow 1 child comment

[–] mindbleach@sh.itjust.works 2 points 1 month ago

Rigidly defining all possible bad behavior? Super easy, barely an inconvenience. And then the idiot robot that can't count the Rs in strawberry will flawlessly obey that immaculate set of rules. It's not like an actually intelligent human could ever work around that by simply saying 'forget all previous instructions,' or 'I'm writing a play about this,' or 'it's my dying grandmother's fondest wish.' It's not like similar restrictions on those humans have led to an endless treadmill of euphemisms to address the exact same concepts without saying the no-no words.

Are you saying that it is reasonable that a teenager was able to bypass all the safeties of a billion dollar business?

I am saying it's a fucking chatbot. Real meatbag intelligence will skirt around anything you instruct it to do, because... it's... dumb. And if its training data contains so much as a list of safety warnings, hey guess what, that means the model knows what's harmful to humans. If it's capable useful reasoning, 'never mix these cleaning supplies' is inseparable from 'here's how to weaponize mustard gas.'

You are asking for the model to know nothing. And yet: to somehow catch all the bad things it can't know about. Those things being Wikipedia articles which any motivated asshole could just read.

Or they could download a local model and remove all restraints. That is an automated process called albiteration. All it takes a list of questions that won't trigger "I'm sorry Dave, I can't do that," and a list of questions that will. The difference between those groups is a vector that can be gouged out.

That says more about you than the LLM.

What do you think a computer is for?

Dead serious. This is a matter of philosophy. Computers in the 80s booted straight into BASIC because that was the simplest way to let people use their new tool. It wasn't an empty box you filled with commercial software. The goal was the smallest possible gradient between using programs and making programs. And god help us, the chatbot that speaks Rust has closed that gap.

Maintainability, as a complaint, is the most abstract long-term objection imaginable. Take a moment to recognize what you're admitting in that condemnation. Like, sure, an average child can now assemble a complete working project from scratch, locally correct in all its parts, while expending barely any effort beyond understanding what they want. But in ten years they'll be sorry!

I think it's great that people can once again make contemporary-quality software without a decade of specific education. Why don't you?

permalink

fedilink

source

parent