Comment by Sharlin
6 hours ago
Problem is people seriously believe that whatever GPT tells them must be true, because… I don't even know. Just because it sounds self-confident and authoritative? Because computers are supposed to not make mistakes? Because talking computers in science fiction do not make mistakes like that? The fact that LLMs ended up having this particular failure mode, out of all possible failure modes, is incredibly unfortunate and detrimental to the society.
Last year I had to deal with a contractor who sincerely believed that a very popular library had some issue because it was erroring when parsing a chatgpt generated json... I'm still shocked, this is seriously scary
"SELECT isn't broken" isn't a new advice, and it exists for a reason.
My boss says it's because they are backed by trillion dollar companies and the companies would face dire legal threats if they did not ensure the correctness of AI output.
Point out to your boss that trillion dollar companies have million dollar lawyers making sure their terms of service put all responsibility on the user, and if someone still tries to sue them they hire $2000/hour litigators from top law firms to deal with it.
Your boss sounds hilarious naive to how the world works.
In a lot of ways he is, despite witnessing a lot of how the sausage is made directly. Honestly, I think at at least half of it is wanting to convince himself that the world still functions in ways that make sense to him rather than admit that it's mostly grifters grifting all the way down.
6 replies →
This is a good heuristic, and it's how most things in life operate. It's the reason you can just buy food in stores without any worry that it might hurt you[0] - there's potential for million ${local currency} fines, lawsuits, customer loss and jail time serving as strong incentive for food manufacturers and vendors to not fuck this up. The same is the case with drugs, utilities, car safety and other important aspects of life.
So their boss may be naive, but not hilariously so - because that is, in fact, how the world works[1]! And as a boss, they probably have some understanding of it.
The thing they miss is that AI fundamentally[2] cannot provide this kind of "correct" output, and more importantly, that the "trillion dollar companies" not only don't guarantee that, they actually explicitly inform everyone everywhere, including in the UI, that the output may be incorrect.
So it's mostly failure to pay attention and realize they're dealing with an exception to the rule.
--
[0] - Actually hurt you, I'm ignoring all the fitness/healthy eating fads and "ultraprocessed food" bullshit.
[1] - On a related note, it's also something security people often don't get: real world security relies on being connected - via contracts and laws and institutions - to "men with guns". It's not perfect, but scales better.
[2] - Because LLMs are not databases, but - to a first-order approximation - little people on a chip!
2 replies →
And just how many rs does your boss think are in strawberry?
If only every LLM-shop out there would put disclaimers on their page that they hope absolve them of the responsibility of correctness, so that your boss could make up his own mind... Oh wait.
I think people's attitude would be better calibrated to reality if LLM providers were legally required to call their service "a random drunk guy on the subway"
E.g.
"A random drunk guy on the subway suggested that this wouldn't be a problem if we were running the latest SOL server version" "Huh, I guess that's worth testing"
There’s a non-zero number of people who would get a chuckle out of a browser extension at replaces every occurrence of LLM or AI with a random drunk guy on the subway .
People's trust on LLM imo stems from the lack of awareness of AI hallucinating. Hallucination benchmarks are often hidden or talked about hastily in marketing videos.
I think it's better to say that LLMs only hallucinate. All the text they produce is entirely unverified. Humans are the ones reading the text and constructing meaning.
[flagged]
To quote Luke Skywalker: Amazing. Every word of what you just said is wrong.
3 replies →
> It's just a database. There is no difference in a technical sense between "hallucination" and whatever else you imagine.
It's like a JPEG. Except instead of lossy compression on images that give you a pixel soup that only vaguely resembles the original if you're resource bound (and even modern SOTA models are when it comes to LLMs), instead you get stuff that looks more or less correct but just isn't.
This comes from not having a specific area or understanding, if you ask it about an area you know well, you'll see.
I get what you're saying but I think it's wrong (I also think it's wrong when people say "well, people used to complain about calculators...").
An LLM chatbot is not like querying a database. Postgres doesn't have a human-like interface. Querying SQL is highly technical, when you get nonsensical results out of it (which is most often than not) you immediately suspect the JOIN you wrote or whatever. There's no "confident vibe" in results spat out by the DB engine.
Interacting with a chat bot is highly non-technical. The chat bot seems to many people like a highly competent person-like robot that knows everything, and it knows it with a high degree of confidence too.
So it makes sense to talk about "hallucinations", even though it's a flawed analogy.
I think the mistake people make when interacting with LLMs is similar to what they do when they read/watch the news: "well, they said so on the news, so it must be true."
Billions of dollars of marketing have been spent to enable them to believe that, in order to justify the trillions of investment. Why would you invest a trillion dollars in a machine that occasionally randomly gave wrong answers?
I don't remember exactly who said it, but at one point I read a good take - people trust these chatbots because there's big companies and billions behind them, surely big companies test and verify their stuff thoroughly?
But (as someone else described), GPTs and other current-day LLMs are probabilistic. But 99% of what they produce seems feasible enough.
I think in science fiction it’s one of the most common themes for the talking computer to be utterly horribly wrong, often resulting in complete annihilation of all life on earth.
Unless I have been reading very different science fiction I think it’s definitely not that.
I think it’s more the confidence and seeming plausibility of LLM answers
People are literally taking Black Mirror storylines and trying to manifest them. I think they did a `s/dys/u/` and don't know how to undo it...
In terms of mass exposure, you're probably talking things like Cmdr Data from Star Trek, who was very much on the 'infallible' end of the fictional AI spectrum.
Sure, but this failure mode is not that. "AI will malfunction and doom us all" is pretty far from "AI will malfunction by sometimes confabulating stuff".
The stories I read had computers being utterly horribly right, which resulted in attempts (sometimes successful) at annihilate humanity.