- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
You must log in or register to comment.
If scientists made AI, then it wouldn’t be an issue for AI to say “I don’t know”.
But capitalists are making it, and the last thing you want is it to tell an investor “I don’t know”. So you tell it to make up bullshit instead, and hope the investor believes it.
It’s a terrible fucking way to go about things, but this is America…
It’s got nothing to do with capitalism. It’s fundamentally a matter of people using it for things it’s not actually good at, because ultimately it’s just statistics. The words generated are based on a probability distribution derived from its (huge) training dataset. It has no understanding or knowledge. It’s mimicry.
It’s why it’s incredibly stupid to try using it for the things people are trying to use it for, like as a source of information. It’s a model of language, yet people act like it has actual insight or understanding.
Imagine searching your computer for a PDF named “W2.2026”…
Would you rather the computer tell you it’s not in the database? Or would you prefer a random PDF displayed with the title “W2.2026”?
This isn’t a new problem.
You’re getting hung up on “know” instead “has relevant information in it’s database and can access it”.
But besides all that and the other things you got wrong:
It’s still about capitalism for the reasons I just said
You do not understand how these things actually work. I mean, fair enough, most people don’t. But it’s a bit foolhardy to propose changes to how something works without understanding how it works now.
There is no “database”. That’s a fundamental misunderstanding of the technology. It is entirely impossible to query a model to determine if something is “present” or not (the question doesn’t even make sense in that context).
A model is, to greatly simplify things, a function (like in math) that will compute a response based on the input given. What this computation does is entirely opaque (including to the creators). It’s what we we call a “black box”. In order to create said function, we start from a completely random mapping of inputs to outputs (we’ll call them weights from now on) as well as training data, iteratively feed training data to this function and measure how close its output is to what we expect, adjusting the weights (which are just numbers) based on how close it is. This is a gross simplification of the complexity involved (and doesn’t even touch on the structure of the model’s network itself), but it should give you a good idea.
It’s applied statistics: we’re effectively creating a probability distribution over natural language itself, where we predict the next word based on how frequently we’ve seen words in a particular arrangement. This is old technology (dates back to the 90s) that has hit the mainstream due to increases in computing power (training models is very computationally expensive) and massive increases in the size of dataset used in training.
Source: senior software engineer with a computer science degree and multiple graduate-level courses on natural language processing and deep learning
Btw, I have serious issues with both capitalism itself and machine learning as it is applied by corporations, so don’t take what I’m saying to mean that I’m in any way an apologist for them. But it’s important to direct our criticisms of the system as precisely as possible.
You don’t seem to understand. There is no database.
Uh, I understand the sentiment, but the model doesn’t know anything. And it’s legit really hard to differentiate between factual things and random bullshit it made up.
Was gonna say, the AI doesn’t make up or admit bullshit, its just a very advanced a prediction algorithm. It responds with what the combination of words that is most likely the expected answer.
Wether that is accurate or not is part of training it but you’ll never get 100% accuracy to any query
If it can name what the most likely combination is, couldn’t it also know how likely that combination of words is?
It’s not actually deciding anything, the AI thinking is marketing fluff really. But yes that’s called confidence rating and it does. But at the scale of something like chatgpt that uses a snapshot of the entire internet and is non mutable there’s no way to train it for every possible question. If you ask about a topic 99% of the internet gets wrong it’ll respond the wrong thing with 99% confidence
If it has been trained using questionable sources, or if it’s training data includes sarcastic responses (without understanding that context), it isn’t hard to imagine how confidently wrong some of the responses could be.
Yeah, no one can make it say “I don’t know” because it is not really AI. Business bros decided to call it that and everyone smiled and nodded. LLMs are 1 small component (maybe) of AI. Maybe 1/80th of a true AI or AGI.
Honestly the most impressive part of LLMs is the tokenizer that breaks down the request, not the predictive text button masher that comes up with the response.
Honestly the most impressive part of LLMs is the tokenizer that breaks down the request, not the predictive text button masher that comes up with the response.
Yes, exactly! It’s ability to parse the input is incredible. It’s the thing that has that “wow” factor, and it feels downright magical.
Unfortunately, that also makes people intuitively trust its output.
It “knows” as in it has access to the information and the ability to provide the right info for the right context.
Any part of that process the AI can just “bullshit” and fills in the gaps with random stuff.
Which is what you want when it’s “learning”. You want it to try so it’s attempt can be rated, and the relevant info added to its “knowledge”.
But when consumers are using it, you want it to say “I can’t answer that”. But consumers are usually stupid and will buy/use the one that says “I can’t answer that” the least.
And it’s legit really hard to differentiate between factual things and random bullshit it made up.
Which is why AI should tell end users “I don’t know” more often.
It “knows” as in it has access to the information and the ability to provide the right info for the right context.
It doesn’t, though, any more than you have access to the information in a pile of 10 million shredded documents.
Right, in this case that we’re talking about…
Do you not understand how “answer unavailable” is a better answer than taking a small percent of strips of paper at random and filling in the rest with words that sound relevant?
It’s like a mad libs
taking a small percent of strips of paper at random and filling in the rest with words that sound relevant?
It’s like a mad libs
Right. They’re text generators. That’s the technology. It can’t do what you’re demanding because that’s not how it works. LLMs aren’t magic answer machines. They don’t know when to say “answer not available”. They don’t know what they’re being asked. They don’t know anything.
You know that answer unavailable is better because you have real intelligence, an LLM is just some mathematical functions so it can’t do that. If it could it would be getting much closer to actually being AI.
Which is why AI should tell end users “I don’t know” more often.
If you feel this is a simple solution, I strongly suggest you write up exactly how you do this and make yourself a billion dollars.
This has nothing to do with scientists vs capitalists and everything with the fact that this is not actually “AI”. Someone called it T9 (word prediction) on steroids and I find that much more fitting with how those LLMs work. It just mimics the way humans talk, but it doesn’t actually converse intelligently or actually understands context - it just looks like it does, but only if you take it at face value and don’t look deeper into it.
Removed by mod
Removed by mod
It’s just short for automatic transmission, opposed to manual transmission. I think Americans call manual cars sticks though. But they’re not sticks, because sticks are wood and cars are almost always metal. Not metal like the music though.
Edit - thinking on it you could play metal through the car stereo though.
I know the difference between an automatic & manual car & transmission. The analogy just doesn’t make sense, because when you say “automatic / manual car” you’re still referring to something within the car, the transmission system - you’re not actually calling the car to be “automated” or whatever. Calling LLMs “AI” however is nothing but a misnomer and that analogy simply does not compare at all.
Removed by mod
Removed by mod
It is made by scientists. The problem is that said scientists are paid by investors mostly, or by grants that come from investors.
And by the time the system can actually research the facts, the internet is so full of LLM generated nonsense neither human or AI can verify the data.
Just ask ChatGPT what it thinks for some non-existing product and it will start hallucinating.
This is a known issue of LLMs and DL in general as their reasoning is a black box for scientists.
It’s not that their reasoning is a black box. It’s that they do not have reasoning! They just guess what the next word in the sentence is likely to be.
No surprise, and this is going to happen to everybody who uses neural net models for production. You just don’t know where your data is, and therefore it is unbelievably hard to change data.
So, if you have legal obligations to know it, or to delete some data, then you are deep in the mud.
Stop asking a language model for accurate information and problem solved. ChatGPT is not supposed to be a knowledge bank, that’s purely incidental for the amount of training data.
Stop asking a language model for accurate information and problem solved
Hey chatgpt, when did jol’s wife get pregnant and by whom?
/s
