Most people agree that there are legitimate applications for LLM technology, like in the medical field. This comes up every single thread.
Then you have trillions of dollars of investment on the promise of eroding worker power, and letting capital enunciate itself without labour. That’s what it’s essentially all for.
The “legitimate” uses of the technology are not even a passing concern for investors except insofar as they can point to it to excuse their grift.
Rightrightright, I mean they can help with that stuff by churning though large sets of data or by statistical analysis to look for early signs of cancer, etc.
I’m not a researcher and I don’t use them myself, but I know a thing or two about how LLMs work. Could be you could school me though if you know a lot.
I’m just saying the models used in medical imaging and protein folding, to use your examples, are not LLMs, but very different dedicated medical imaging models or protein folding models. Still falls under the overarching umbrella of AI/ML, but not really any more relevant to language models. LLMs also aren’t a very relevant tool for big data or statistical analysis. Even with an agent harness, they could be used to write scripts that a human could also write, but couldn’t directly do any sort of reliable statistical analysis.
Oh OK I wasn’t aware of that. Actually I wondered as I was writing my comment how an LLM would compare to more tightly fitted model.
Well there goes my one example. Is there anything they’re legitimately good for? I’ll admit they seem good at finding “fuzzy” connections, and their text and image generation capabilities are impressive even if they aren’t my cup of tea—mostly because of the externalities.
They’re good for coding simple things for yourself if you don’t want to learn to code or spend hours making something you will use once. Kinda like fast food.
You could argue I can learn to cook better and spend the time to do so. But I don’t always want to and fast food is good enough. Same with llm. I can ask it to make me a Web app that does X and have it in seconds. All the usual worries about bugs and security don’t apply as strongly to an app I make for just me that does just one thing.
Like say I’m at a takeaway shop with multiple people and the only menu is the board behind the counter. I take a couple photos, send them to the llm and say make me a menu app that I can add and remove items and it tracks the price. I will have it in under a minute and we can stop trying to read the board and block the counter and have the cashier patiently staring at us while we browse it in our own time and know exactly what the total will be.
Now you can quibble over whether that’s “genuinely useful” or not. But it is to me. And I make tons of these simple one use apps for all kinds of things. And even if I was the best coder on earth, it still wouldn’t be practical to hand make the same thing on the fly. Also hyper specific Firefox add-ons that don’t exist because they’re too specific to me and I didn’t want to learn to make Firefox add-ons just to have them.
Even on the larger end, I wanted a lucid dream journal app and at the time the ones on the play store largely pay walled the best features and even pay walled exporting your journal. I didn’t like that. So I had llm make me a basic one with the features I wanted, everything stored as txt files with export/import and ability to back up to drop box by oauth into a dedicated folder that can’t touch the rest of my Dropbox. Even if llm disappeared and the app broke and I couldn’t fix it I still have my entire journal as plain text.
So yeah. I find it useful. Because the alternative isn’t me learning to code. It’s these things not existing.
You say most people believe this, but I see many people every day for weeks now claiming Linux is vibecoded now. Why? Three stories: 1) months ago LLM use was not outright banned in kernel development 2) Linus admitted to using LLMs in an experimental branch to find and patch bugs 3) a recent shift in attitude happened, toward LLM-driven PRs because there were too many, and the reaction from Linus and the other maintainers was to raise the bar for submitting PRs to reduce the chance of slop wasting people’s time.
All of this was a series of level headed responses and course corrections which at no time was indicative of anything more than LLMs are useful sometimes. Yet people all over Lemmy are saying shit like “Linux is all in on vibecoding”. Besides that issue, I’ve seen and been involved in many conversations where idiots will say that any use of AI is super unethical and unacceptable. The binary thinking on this issue is strong on here.
Ah well I agree I would Linux love maintainers to take a more principled stance against LLM contributions given all the negative externalities of the technology as it exists today.
there’s a strong case to be made that there is no way to use the current “frontier” models ethically because of how they were trained and who their patrons are. I would love to see that acknowledged.
By that logic most media consumption is unethical. Almost anything out of a major studio is made with the abuse of, underpayment of, and often even outright theft from small artists. I think a much better solution than not using them is to take all the money they made from it and redistribute it as universal basic income, then make the models 100% free to use. Cat’s outta the bag, at this point I just wanna see the little guy get paid.
By that logic most media consumption is unethical. Almost anything out of a major studio is made with the abuse of, underpayment of, and often even outright theft from small artists.
Now we’re on the same track. My goal is if we can’t solve these problems entirely, at the least not contribute to the further exploitation and enclosure of our commons.
This is one of the reasons I think copyright is a dead end — it’s a tool of the enemy.
You thought you were being subtle but if your response to an extremist viewpoint is “yes they make a good point” you’re not getting it but you do seem to want to definitely take a side.
“I’m just being objective but your point isn’t worth addressing while the thought process you take issue with is totally valid”
I’m not trying to be subtle, I’m trying to be open. Clearly we’ve each got our opinions and I think I’ve been very upfront with mine. You’re right I don’t think being against this technology qualifies as an “extremist” position anymore then being against cloud computing or Nvidia are extreme takes.
Anyway you didn’t even try me. I’m elsewhere in this thread talking about the legitimate uses for LLMs. Not everything is as black vs white, clanker vs luddite, as you’re making it.
Yes being so against a technology that you will spread lies about it makes one an extremist. Siding with that… Whatever. Have fun with that. I won’t be seeing anymore of this from you.
Most people agree that there are legitimate applications for LLM technology, like in the medical field. This comes up every single thread.
Then you have trillions of dollars of investment on the promise of eroding worker power, and letting capital enunciate itself without labour. That’s what it’s essentially all for.
The “legitimate” uses of the technology are not even a passing concern for investors except insofar as they can point to it to excuse their grift.
What is the legitimate use for LLMs in the medical field?
Like medical research, discovering drugs, folding proteins. There’s also evidence they can be used to help analyse x-rays.
LLMs don’t do those things. Do you know what an LLM is?
Rightrightright, I mean they can help with that stuff by churning though large sets of data or by statistical analysis to look for early signs of cancer, etc.
I’m not a researcher and I don’t use them myself, but I know a thing or two about how LLMs work. Could be you could school me though if you know a lot.
I’m just saying the models used in medical imaging and protein folding, to use your examples, are not LLMs, but very different dedicated medical imaging models or protein folding models. Still falls under the overarching umbrella of AI/ML, but not really any more relevant to language models. LLMs also aren’t a very relevant tool for big data or statistical analysis. Even with an agent harness, they could be used to write scripts that a human could also write, but couldn’t directly do any sort of reliable statistical analysis.
Oh OK I wasn’t aware of that. Actually I wondered as I was writing my comment how an LLM would compare to more tightly fitted model.
Well there goes my one example. Is there anything they’re legitimately good for? I’ll admit they seem good at finding “fuzzy” connections, and their text and image generation capabilities are impressive even if they aren’t my cup of tea—mostly because of the externalities.
They’re good for coding simple things for yourself if you don’t want to learn to code or spend hours making something you will use once. Kinda like fast food.
You could argue I can learn to cook better and spend the time to do so. But I don’t always want to and fast food is good enough. Same with llm. I can ask it to make me a Web app that does X and have it in seconds. All the usual worries about bugs and security don’t apply as strongly to an app I make for just me that does just one thing.
Like say I’m at a takeaway shop with multiple people and the only menu is the board behind the counter. I take a couple photos, send them to the llm and say make me a menu app that I can add and remove items and it tracks the price. I will have it in under a minute and we can stop trying to read the board and block the counter and have the cashier patiently staring at us while we browse it in our own time and know exactly what the total will be.
Now you can quibble over whether that’s “genuinely useful” or not. But it is to me. And I make tons of these simple one use apps for all kinds of things. And even if I was the best coder on earth, it still wouldn’t be practical to hand make the same thing on the fly. Also hyper specific Firefox add-ons that don’t exist because they’re too specific to me and I didn’t want to learn to make Firefox add-ons just to have them.
Even on the larger end, I wanted a lucid dream journal app and at the time the ones on the play store largely pay walled the best features and even pay walled exporting your journal. I didn’t like that. So I had llm make me a basic one with the features I wanted, everything stored as txt files with export/import and ability to back up to drop box by oauth into a dedicated folder that can’t touch the rest of my Dropbox. Even if llm disappeared and the app broke and I couldn’t fix it I still have my entire journal as plain text.
So yeah. I find it useful. Because the alternative isn’t me learning to code. It’s these things not existing.
You say most people believe this, but I see many people every day for weeks now claiming Linux is vibecoded now. Why? Three stories: 1) months ago LLM use was not outright banned in kernel development 2) Linus admitted to using LLMs in an experimental branch to find and patch bugs 3) a recent shift in attitude happened, toward LLM-driven PRs because there were too many, and the reaction from Linus and the other maintainers was to raise the bar for submitting PRs to reduce the chance of slop wasting people’s time.
All of this was a series of level headed responses and course corrections which at no time was indicative of anything more than LLMs are useful sometimes. Yet people all over Lemmy are saying shit like “Linux is all in on vibecoding”. Besides that issue, I’ve seen and been involved in many conversations where idiots will say that any use of AI is super unethical and unacceptable. The binary thinking on this issue is strong on here.
Ah well I agree I would Linux love maintainers to take a more principled stance against LLM contributions given all the negative externalities of the technology as it exists today.
there’s a strong case to be made that there is no way to use the current “frontier” models ethically because of how they were trained and who their patrons are. I would love to see that acknowledged.
By that logic most media consumption is unethical. Almost anything out of a major studio is made with the abuse of, underpayment of, and often even outright theft from small artists. I think a much better solution than not using them is to take all the money they made from it and redistribute it as universal basic income, then make the models 100% free to use. Cat’s outta the bag, at this point I just wanna see the little guy get paid.
Now we’re on the same track. My goal is if we can’t solve these problems entirely, at the least not contribute to the further exploitation and enclosure of our commons.
This is one of the reasons I think copyright is a dead end — it’s a tool of the enemy.
The models aren’t making money and they aren’t going to, though.
So you are on the side of binary thinking, in the end. Cool.
What makes you think that? I am on no side. I’m just trying to be objective. Everything I said is factually borne out afaik.
Also
Lol now you’re just playing with me, very funny ;p
You thought you were being subtle but if your response to an extremist viewpoint is “yes they make a good point” you’re not getting it but you do seem to want to definitely take a side.
“I’m just being objective but your point isn’t worth addressing while the thought process you take issue with is totally valid”
I’m not trying to be subtle, I’m trying to be open. Clearly we’ve each got our opinions and I think I’ve been very upfront with mine. You’re right I don’t think being against this technology qualifies as an “extremist” position anymore then being against cloud computing or Nvidia are extreme takes.
Anyway you didn’t even try me. I’m elsewhere in this thread talking about the legitimate uses for LLMs. Not everything is as black vs white, clanker vs luddite, as you’re making it.
Yes being so against a technology that you will spread lies about it makes one an extremist. Siding with that… Whatever. Have fun with that. I won’t be seeing anymore of this from you.
Lies, like… ?