Should AI companies be liable for LLM output?
(media.kotakuinaction2.win)
Comments (48)
sorted by:
People who outsource their thinking believe they should be exempt from the consequences of doing so. You'll have to imagine my surprise, because it doesn't exist.
Should PlaySkool be held liable if you write "fuck" on a Mr Spell? Should Proctor and Gamble be held liable if you try to drink bleach?
Let the idiots die.
i know it's cathartic to say stuff like this, but watch that someone you know or love doesn't become one of those idiots who dies. there's a whole lot of people for whom the difference between a decent life and ruin is just what vices they're given easy access to - you probably know a few - and that means that with just a little empathy and effort you can skew the results a lot towards more decent lives than ruined ones.
Can happen, has happened. More than once. My opinion remains the same.
If you are dumb enough to kill yourself because a silly little toy told you to, then odds are it's not the only thing that would have ended your life eventually. And you can't say that it's the mere presence of a vice either, in my opinion. If you drink yourself to death because alcohol exists, you were going to fall down some hole somewhere at some point in your life.
For me, it's pragmatic. My children will have a better life, if the kind of people who got lawn darts banned weren't in it.
Third panel Anakin.
This is why commercial chatbots refuse to give out any of the useful information they contain.
Everyone who answers this question in the affirmative needs to be banned from ever interacting with a LLM, as well as several other technologies such as the internet and motor vehicles.
Reminder that every "survey" or "study" is almost entirely conducted on college campuses or cities near them. Meaning that the sample size is overwhelmingly favored towards young Lefties.
So these things are more of a reflection of that specific demographic instead of "US Voters" in general.
I never thought of this. But I was surveyed many times while at uni and almost never since…
a few dollars for my time was more appealing back then
The money is always good, mostly because its easy and free. And most college kids are gonna be there anyway, so why not.
But almost all of these things are operated out of such locations, so they aren't going to travel far if they don't have to and technically a campus isn't Leftist only so they can pretend its diverse.
The only alternatives you ever see are phone surveys and online polls, both of which have such a self selection bias of who the fuck would answer this that they are useless for anything other than social engineering headlines.
I recall one college Psych study where they paid me something like 15 or 25 bucks to get drunk.
God damn man, that's an absurd level of payment for one of these. We were taught to give like 5$ back in the day.
It was a couple of hours of time commitment back in the aughts where you did a before & after timed computer test under the influence.
I remember getting tipsy on the standardized clinical alcohol. Then flirting with the graduate level test administer girl. After the experiment was over, she kept getting me to do the breathlyzer over & over again waiting for me to sober up so she could yeet me.
I suppose for a multi hour inclusion that makes more sense, that's also a study that isn't going to get huge sample sizes so their budget for paying can dole out more.
And your attitude into it is another reason why these things are so weak. Very few people are taking them serious, and often in it for the hot Psych undergrads running it, so their answers are usually pretty underthought or unrevealing.
What incentive do survey companies have to ever actually ask any person any questions? They are paid to produce a result. No one asks about the historical reputation of the surveyor—we just consume the results as fact.
Just the fact that they need to do enough work that it looks like they did something, enough to satisfy their sponsor. Otherwise whoever paid them could have just paid a random PhD to attach their name to a made up quote entirely, and we aren't quite that far gone yet most of the time.
And a lot of the time these things are being done by grad students or research assistants, so there is a grade being held over their heads to at least conduct something.
You are right about the rest though, no one reads past the headline and the chart. Because if they did it would be comically obvious how weak 90% of these are in telling you anything.
No it's your fault for not using your brain and cross-checking as if it were any other source.
Devil's advocate: using your brain and cross-checking is based on a ton of knowledge you absorbed before the information landscape turned to dogshit. When they "use their brains" They're not cross-checking AI against a reputable textbook they read authored by people who dedicated their careers to factual accuracy. They're cross-checking AI against a 24-hour news cycle, TikTok influencers, and activism masquerading as science.
At what point does functional illiteracy become their own fault?
Honestly, that's a tough one. We had it relatively easy. There was the usual mainstream media propaganda, but the Internet of the time offered a place for honest conversation. That's scarce now and under constant threat. I don't see anything on the horizon replacing it unless the Euro censorship pushes drive everyone to the darkweb, which would probably be the best outcome. Not to go full black-pill, but think about it:
What are the odds that someone on the younger side is functionally literate even if they make an above-average effort to be informed? And it's by design, outside of their control. What sane control information do they have to cross-check things against? It's bleak. Unless you have a root of trust, you're going to be cross-checking garbage against garbage. If you get lucky enough, maybe you accidentally come across enough truth that one day it clicks and upends everything you know, but I don't think that's especially common.
Even in the Soviet Union people could still properly read, and that was a place where you couldn't tell jokes without going to jail. I don't think the environment of the times is much excuse.
This isn't something that has to be handed to you. It's a matter of effort. The true problem we have is laziness and complacency
I think we're arguing past one another. I'm not arguing that we don't have a laziness and complacency issue. I'm saying that it's become more difficult to separate the wheat from the chaff. Per effort, the achieved literacy is less. The laziness and complacency compounds with it.
Sure. Now increase the number of books they had access to by a factor of 10,000 but make all the new ones a mix of propaganda, AI-generated nonsense, and bad fanfic.
I'm not trying to absolve them of their lack of effort. I'm just saying they're in a markedly different situation than people born 20+ years prior.
And I'm pointing out that in living memory there existed a tyrannical society whereby nothing existed except state dictated information and we still had higher quality people than we do now. I lived through this tyrannical society myself, although the tail end of it thankfully for me. Shit was worse than it is now overall.
The sludge that is the internet isn't the proximate cause here imo.
Yes it's shitty. Yes the internet is a pale shade of what it once was and most things of use have been wrung out of it and replaced with junk.
But that's not why idiots try to cook chicken in Pepto Bismal.
I think being flooded with garbage from a thousand different angles may have a worse impact than having to differentiate between one flavor of propaganda and everything else. Like filtering out a specific, recognizable signal vs. being jammed by noise. How often was it that you couldn't tell if a given piece of information was state-dictated or not?
I think we agree that laziness, complacency, and a thoroughly poisoned well are all problems. Not much point in debating the exact magnitude of each.
And to your point, OP says registered US voters so these idiots are at least 18.
I want to point out that we have been deliberately teaching people how to read incorrectly for a long time now. The increasing illiteracy rates are a feature, not a bug for the elite.
If–from birth–you’re never trained to read, how can it be your fault?
Adults can't learn? Teenagers can't learn?
Nobody can achieve anything of their own accord, it has to be handed to you?
Apparently they can’t even conceive of the idea. That’s what happens when you lower your societal standards to the level of a species that never invented the written word.
Of course. "Democracy" can have no other outcome.
Once you decide that the bar is "still breathing", then that is always and inevitably going to happen.
Its not that they can't learn, its that they literally need to unlearn the incorrect methods they were taught. Which is substantially harder to do then simply learning a new skill.
These are the same people who want to hold gun manufacturers liable for shootings. They really are that fucking stupid.
But hold on. Can we hold reddit users responsible for those three categories as well? The WSB lawsuits alone would be something to behold. Actually, yeah. This is a great precedent. Everyone who bought GME weeklies should be able to get full restitution from Reddit. Not the commenters, but Reddit itself.
"Mommy AI, tell me what to do."
"Daddy government, make AI be nice to me"
The world is full of absolute children, with no idea of how to get what they want, just the egotistical compulsion that they should get it.
Chatbots should tell everyone on earth to kill themselves as filter for the weak minded and to strengthen the resolve of all who remain.
What we should focus on is the deliberate shitting up of every social media platform with automated content. Altman acts like there's no way to solve this but bitch just check the logs. You're producing all the text, just check the logs and ban people who are making spam emails and automated tweets.
Watch them use the child pron laws to get the legal framework set up.
Just remember that every AI trained on facts, logic, reason, and evidence became super right wing and hilariously racist.
They should be held liable for misrepresenting the capabilities of the LLM. If someone rents out an LLM as a Magic 8-Ball and someone else rents out a wrapper for that LLM as a financial advisor, the latter should be held accountable for their fraud but the former should not.
I think you’re right where the responsibility really lies.
Also - Every time I use a LLM it clearly says “llms can make mistakes and hallucinate”
sauce :
https://archive.is/MTgSD
They should be held liable if they're run by my enemies and holding them liable is a way I can hurt them. Otherwise I don't really care.
So, if it draws a picture, it's breaking copyright of thousands of artists simultaneously, and is liable for damages, but if it says something wrong, it's wholly responsible, and is liable for damages. Got it.
I agree with many of the responses here, but at the same time, these multi-hundred-billion dollar systems were designed to be social engineering tools that undermine regular people.
We're at the point right now where the top search engine results are all "AI" slop instead of factual references. This means it's getting more and more time-consuming to tease out what's real and what's fake. At some point, we may be past some AI event horizon and be completely unable to discern what's fabricated and what's real by ourselves in a reasonable amount of time. In a situation like that, should LLMs not be held responsible for their suggestions?
Somewhat of an aside, but whatever you may think of AI and LLMs, their intended purpose is to make obsolete and remove what tiny power and influence the little people have in corporations and re-concentrate it in the elites in charge. I'm not thrilled at the idea of absolving them of liability.
If people expect some inherent God level truth & accuracy from an LLM, then they are complete idiots. This is why we can't have nice stuff.
“Negative outcomes” is very broad, so that terminology makes me suspicious. What if the LLM gives you accurate information, but you’re a retard and use that accurate information poorly? I do think that it’s obviously bad if the LLM is giving outright lies out, although given the way they hallucinate sometimes on basically any topic, I’m not convinced that that’s something the company can be blamed for so much as a ubiquitous, unavoidable flaw in the early versions of an emerging technology.
One of the scariest things about LLMs is how many people treat an obviously fallible, obviously open to bias algorithm as an arbiter of truth. I’ve seen people do things like respond to twitter posts about bias in a Wikipedia article with “@grok, is there bias in this article?” of course, Grok ingests loads of training data from Wikipedia itself and the various MSM sources they cite. There’s no guarantee Grok wouldn’t just have that same bias, and yet people assume it is an authoritative source. That’s an obvious issue that is worrying at a societal level, but how do you legislate for “there’s a really amazing technology with a lot of potential, but many people are dumb?”
I'm going to go against the grain here and say it depends.
If you, the ai maker, claim your ai is a medical guru who will give real and proper clinical advice then you absolutely should be held liable if your AI gives wildly incorrect directions.
If you do not make claims such as this, then you should not be held liable.
This is why I hate surveys like this. Because it is very easy to interpret the question in different ways.
Yes they should. Why? Cuz fuck em that's why.
Everyone here is held liable for crimes they commit, even if they're ignorant of the law. But when AI gives harmful advice, the multi-billion dollar company that built it will never be punished in any meaningful way.