A papyrus scroll that was burned and carbonized when Mount Vesuvius erupted almost 2,000 years ago has been virtually unrolled and partially deciphered with the help of artificial intelligence.
Well when they said the Pentagon was using AI to help coordinate the attacks on Iran, I simply assumed it was some expert system with a machine learning component. But no, it turns out they were using Grok, an actual LLM chatbot. Which is fucking crazy.
So sometimes the assumption goes the other way as well.
Well, having the tools and knowing how best to use them are two different things. The fact they used an LLM to determine how to use the tools is indicative of them not being overly adept at using them. Also, the fact they got a worse outcome than relying on experts, which they should have easy access to, is typical of situations where LLMs have been used to replace experts. So you’re correct in that this single instance of LLMs doing poorly isn’t a good way to determine if it is useful (in this field), but it follows a common trend of trying to shoehorn LLMs into adjacent fields and failing spectacularly.
On a broader note, the results of a war tend to follow the better leaders. America spent 20 years at war in a large zone with very difficult goals, and easily held their own, even if they couldn’t reach the overall conditions for ultimate success. Now they are in a war with a much smaller field of activity and fairly narrowly defined goals, so should have no trouble building siege conditions to wear the opponent down, yet did it so poorly and somehow attached it to neighboring wars in such a way as to make both the goals harder to achieve and the risks to do so greater. This isn’t the fault of LLMs, but it doesn’t surprise me that someone who would do that would also turn to LLMs to help draft their battle plans after weeding out their more experienced generals.
I doubt the government directly manages military decisions. I’d imagine the people taking these decisions are in that position regardless of which party is at the government.
No there’s an unfortunate thing where AI now means you need a transformer model in some capacity. It’s so bullshit now. I hate these idiotic researchers sometimes.
All AI is actual AI. It doesn’t need to be real intelligence to be artificial, should be obvious. Are you telling me artificial grass shouldn’t be called that because a goat can’t eat it?
Except it is a fairly good artifice of grass. “AI” is not a good artifice of intelligence. GAI would be, but not LLMs or anything else we have today. They aren’t trying to mimic intelligence. They’re trying to mimic the output of intelligence. They don’t think; they reproduce.
Good for what? Looking at? It’s not gonna satisfy a goat.
We only care for the look (“the output”), and we don’t expect more of it, or sell it for more than it is. That’s why it’s not a controversial term for astroturf. It wasn’t controversial for AI either until very recently. In 60 years of AI nobody has split hairs over output of intelligence. It’s justified but weirdly misdirected anger.
Mostly for the thing we care about grass for: looking at and touching. We don’t grow grass to feed goats. We grow grass for the appearance of grass. It serves no practical purpose in most places where grass is grown. Obviously it does have a place in nature, but not where you’re putting artificial grass. The only purpose in those places was touch and appearance. It handles that fine.
AI is a perfectly good term. It’s just an overused one. It was great when it was in academia, and everyone knew what it meant, or in a sci-fi story, and people didn’t think it existed. It’s currently been used as a marketing term. People hear it and think it’s intelligent. It isn’t. The goal of using it for marketing was to create this distortion. They don’t care to inform people of the meaning because the only reason it’s valuable is because of the confusion.
I’d even argue that generative AI is machine learning, except the learning stops when the training does so it’s not learning continuously like ML in the classical sense.
Actual AI, learning models, not large language models hallucinating things.
Indeed this was probably Convolutional Neural Network
This is the prize https://scrollprize.org/
Exactly. The fact that “AI” has now been almost completely associated with LLMs is incredibly frustrating.
So this is more of machine learning than AI?
AI is the larger umbrella term. LLMs, Machine Learning, CNNs, DNNs, RNNs, etc. all fall under that umbrella “AI” term.
Yeah, even pathfinding is “AI” but people associate it only with slop these days.
Like some Dunning Kruger effect sweeping the world.
Well when they said the Pentagon was using AI to help coordinate the attacks on Iran, I simply assumed it was some expert system with a machine learning component. But no, it turns out they were using Grok, an actual LLM chatbot. Which is fucking crazy.
So sometimes the assumption goes the other way as well.
I’d imagine they measured metrics of other automated systems and compared with LLM and found out LLM works best.
It’s not like traditional machine learning models are suddenly perfect and work in all cases.
Looking at the results of the war in Iran, I’d suggest your assumptions are incorrect.
The results of the war do not depend solely on the quality of a single tool or weapon.
The US has better weapons overall and is not winning, thus it is not a way to discriminate whether this technology is useful or not.
Well, having the tools and knowing how best to use them are two different things. The fact they used an LLM to determine how to use the tools is indicative of them not being overly adept at using them. Also, the fact they got a worse outcome than relying on experts, which they should have easy access to, is typical of situations where LLMs have been used to replace experts. So you’re correct in that this single instance of LLMs doing poorly isn’t a good way to determine if it is useful (in this field), but it follows a common trend of trying to shoehorn LLMs into adjacent fields and failing spectacularly.
On a broader note, the results of a war tend to follow the better leaders. America spent 20 years at war in a large zone with very difficult goals, and easily held their own, even if they couldn’t reach the overall conditions for ultimate success. Now they are in a war with a much smaller field of activity and fairly narrowly defined goals, so should have no trouble building siege conditions to wear the opponent down, yet did it so poorly and somehow attached it to neighboring wars in such a way as to make both the goals harder to achieve and the risks to do so greater. This isn’t the fault of LLMs, but it doesn’t surprise me that someone who would do that would also turn to LLMs to help draft their battle plans after weeding out their more experienced generals.
This is the most vibe-coded democratic administration in history. They measured nothing, and fired anyone suggesting otherwise.
I doubt the government directly manages military decisions. I’d imagine the people taking these decisions are in that position regardless of which party is at the government.
That used to be the case, but they are trying to fire anyone who says anything they don’t want to hear.
No there’s an unfortunate thing where AI now means you need a transformer model in some capacity. It’s so bullshit now. I hate these idiotic researchers sometimes.
LLM is “actual AI”. I think the term you may be looking for is “generative AI”.
Correct, but it’s semantics. Most people just think all AI is generative these days, and so I’m trying to differentiate.
Well there is what people think and then there is reality
Yes, but people are reading Lemmy and responding in this thread.
In the old sense of the word, we don’t have an actual AI yet, but it’s true that LLM and AI have become interchangeable.
All AI is actual AI. It doesn’t need to be real intelligence to be artificial, should be obvious. Are you telling me artificial grass shouldn’t be called that because a goat can’t eat it?
No, he’s questioning the intelligence part and not if it’s artificial or not.
Do you question the grass part of artificial grass? It’s obviously not grass.
Except it is a fairly good artifice of grass. “AI” is not a good artifice of intelligence. GAI would be, but not LLMs or anything else we have today. They aren’t trying to mimic intelligence. They’re trying to mimic the output of intelligence. They don’t think; they reproduce.
Good for what? Looking at? It’s not gonna satisfy a goat.
We only care for the look (“the output”), and we don’t expect more of it, or sell it for more than it is. That’s why it’s not a controversial term for astroturf. It wasn’t controversial for AI either until very recently. In 60 years of AI nobody has split hairs over output of intelligence. It’s justified but weirdly misdirected anger.
AI is a perfectly cromulent word for the thing.
Mostly for the thing we care about grass for: looking at and touching. We don’t grow grass to feed goats. We grow grass for the appearance of grass. It serves no practical purpose in most places where grass is grown. Obviously it does have a place in nature, but not where you’re putting artificial grass. The only purpose in those places was touch and appearance. It handles that fine.
AI is a perfectly good term. It’s just an overused one. It was great when it was in academia, and everyone knew what it meant, or in a sci-fi story, and people didn’t think it existed. It’s currently been used as a marketing term. People hear it and think it’s intelligent. It isn’t. The goal of using it for marketing was to create this distortion. They don’t care to inform people of the meaning because the only reason it’s valuable is because of the confusion.
I’d even argue that generative AI is machine learning, except the learning stops when the training does so it’s not learning continuously like ML in the classical sense.
Inference and training are separate in every ML architecture, what are you on about? And yes LLMs are ML, by definition, no need to argue.