Because there are problems out there that you’re not going to feasibly solve on your own.
Think of it this way: do you ever look up anything on the internet? If yes - why did you do that, when you could just find out the information yourself by working from first principles? Likely not only because it’s orders of magnitude faster, but because doing it yourself would probably result in significantly lower quality thank just looking it up on Wikipedia.
LLMs are like search engines on steroids. They really can help you explore problems that have already been described in the literature, so to speak.
But of course that changes how our brains need to be wired. And it’s not necessarily bad. When I was growing up, the adults were really worried that the use of search engines would degrade the ability of children to look things up in the library. And they’re right - I would indeed have a more difficult time going to the library, finding the right shelf, finding the right book, looking through the index, and scanning the pages for the information I was looking for - because that’s a skill that I haven’t practiced because I could just do a web search for the better part of my life. Doing a web search is a different skill.
That’s not to say that this is all fine and just a technological development like any other. I think we do need to be careful about how we use LLMs and when, especially in order to think through problems ourselves. I think that we were already over-reliant on the Internet before LLMs came on the scene, and it will likely get worse now. But the reason that they’re dangerous is because they are in fact really useful tools too.
That would be a good argument if not for the simple fact that ai straight up lies to you.
And its not even a question of using a “good” or “bad” model. We have observed a mathematical constant that shows no matter how large of a model you use, you can’t get below a certain percent of hallucinations.
Its far less common because models got bigger (and more Ressource intensive), but it still happens too often.
Or to phrase it in a different way, would you use a calculator if you knew every 30th answer it gives is wrong?
I see this argument often on Lemmy and I honestly have some trouble empathizing with it. Why should we demand that LLMs would be perfect, when nothing in life is perfect?
Books and scientific reports can, and often have, falsehoods printed in them. All sensors have error margins. A human expert can be convinced that they’re right and still be wrong.
What I’m trying to say is we shouldn’t trust anything blindly. I absolutely agree that it’s frustrating to see so many people trust LLMs blindly, but used as a tool among others I think they’re most likely going to be very effective in a wide variety of fields.
LLMs are out there solving math problems, helping with biology research and fixing latent bugs in widely used open source software. Yes of course LLMs make mistakes, but you can verify their work.
Because there are problems out there that you’re not going to feasibly solve on your own.
Think of it this way: do you ever look up anything on the internet? If yes - why did you do that, when you could just find out the information yourself by working from first principles? Likely not only because it’s orders of magnitude faster, but because doing it yourself would probably result in significantly lower quality thank just looking it up on Wikipedia.
LLMs are like search engines on steroids. They really can help you explore problems that have already been described in the literature, so to speak.
But of course that changes how our brains need to be wired. And it’s not necessarily bad. When I was growing up, the adults were really worried that the use of search engines would degrade the ability of children to look things up in the library. And they’re right - I would indeed have a more difficult time going to the library, finding the right shelf, finding the right book, looking through the index, and scanning the pages for the information I was looking for - because that’s a skill that I haven’t practiced because I could just do a web search for the better part of my life. Doing a web search is a different skill.
That’s not to say that this is all fine and just a technological development like any other. I think we do need to be careful about how we use LLMs and when, especially in order to think through problems ourselves. I think that we were already over-reliant on the Internet before LLMs came on the scene, and it will likely get worse now. But the reason that they’re dangerous is because they are in fact really useful tools too.
That would be a good argument if not for the simple fact that ai straight up lies to you.
And its not even a question of using a “good” or “bad” model. We have observed a mathematical constant that shows no matter how large of a model you use, you can’t get below a certain percent of hallucinations.
Its far less common because models got bigger (and more Ressource intensive), but it still happens too often.
Or to phrase it in a different way, would you use a calculator if you knew every 30th answer it gives is wrong?
I see this argument often on Lemmy and I honestly have some trouble empathizing with it. Why should we demand that LLMs would be perfect, when nothing in life is perfect?
Books and scientific reports can, and often have, falsehoods printed in them. All sensors have error margins. A human expert can be convinced that they’re right and still be wrong.
What I’m trying to say is we shouldn’t trust anything blindly. I absolutely agree that it’s frustrating to see so many people trust LLMs blindly, but used as a tool among others I think they’re most likely going to be very effective in a wide variety of fields.
LLMs are out there solving math problems, helping with biology research and fixing latent bugs in widely used open source software. Yes of course LLMs make mistakes, but you can verify their work.