Transcript
Title text: This is how you all fucking sound
[A smug tech bro wearing a sideways cap, watch, chain around his neck stands in front of a data center by a lake with dead fish. A smoke stack blows pollution into the air]
Tech bro: AI is already here, there’s no going back.
[A smug man in a suit with cigarette in hand stands in a restaurant while two disgruntled diners cough from the smoke]
Suit: Smoking indoors is already here, there’s no going back.
[A smug man in a top hat and suit stands in a factory with two sad and dirty children]
Hat: Child labor is already here, there’s no going back.
[A smug plantation owner stands in front of a field with with two angry slaves]
Plantation owner: The Atlantic Slave trade is already here, there’s no going back.
Yes, but at least at the end of the day you can use nukes to blow stuff up. Presumably your enemies.
If your enemies win the generative AI "arms" race they can use it to, uh...
???
(Yes, I am aware there are military/governmental applications for neural net learning technologies but they're the types of pattern recognition and signals analysis stuff we already do without needing to build a football stadium sized datacenter every 50 miles and burn the entire nation's GDP on electricity generation. Most of the other applications appear to revolve around a regime using it solely to shoot themselves in the foot, e.g. powering a fantasy army of likely to be highly defective murder robots or using it to propagandize at and spy upon their own population in order to ensure a ready supply of destabilizing internal dissent always exists.)
GenAI is really fucking useful for propaganda and disinformation warfare.
That may sound like a compliment to GenAI, it's not.
Unfortunately, the ability to programmatically weaponize false-humanity is real, and really effective.
Which is yet another ethical question that has yet to be widely judged before being deployed everywhere. Just a bunch of greedy, short-sighted exceptionalists.
LLMs are not the final state of AI
But LLMs are not the path to the final state of AI, either. And that's assuming only if — and this is a very big "if" — a true general artificial intelligence can even be created using traditional silicon computing methods in the first place. Blithely assuming that it can be is really rather asking past the sale.
Yep, by design LLM cannot become 'inteligent', you can only make it more believable but it's still copying humans not really thinking by itself. No amount of development or money invested will change that, it's not a pokemon it won't just evolve into something different one day.
To strongman your argument, "LLMs with a supervised training pipeline cannot become intelligent".
RL training pipelines are much more open-ended, and experts still unsure one way or the other if an LLM + RL could lead to intelligence.
And it's worth reiterating, the current crop of generative "AI" is incapable of producing anything new or novel. All it can do is reassemble existing strings, tokens, and patterns in slightly different ways. Innovation can never come from such a machine. That will have to come from a human.
The current push is the notion that "hyperscaling," i.e. throwing even more hardware and space and power and money at the same concept, will magically make it something it isn't. Obviously that's not going to work. It'll allow grifters to make a ton of money over it, though!
Well said.
https://www.scientificamerican.com/article/ai-just-solved-an-80-year-old-erdos-problem-and-mathematicians-are-amazed/
Depends on your definition of novel.
From TFA:
So, it's a monkeys-on-typewriters situation with the computer able to try and reject the hammering of who knows how many square pegs into round holes until it finally arrives at a workable conclusion, which a human has already bested. And we're not allowed to see its homework.
This is categorically failing to set the world on fire, except possibly in the literal sense.
There's value having tedious work done by AI so it can provide inspiration to real people, which is exactly what happened in this case.
Gee, sounds like it's enabling people! The horror.
Things can be useful in the right context without setting the world on fire.
I think LLMs are intelligent. They're at least as intelligent as My pocket calculator, and My calculator is intelligent.
I think you're setting the bar too low. A tiny amount of intelligence is super easy to program.
Not only that, it’s basically eating all the resources that could go into making AGI.
There is nooooo way for companies to invest in actual innovation when they are throwing everything at this dead end.
Then you’re well aware of the massive power that AGI will bring to any nation that can harness it. And no, LLMs alone are not the path, and possibly not the path at all.
Such a fallacy. Anything that falls under the umbrella of machine learning will contribute to future AI. We certainly won’t improve LLMs such that they become AGI, but all of it contributes.
And, whether or not future AI even uses traditional silicon computing is also irrelevant.
What matters is improved understanding of mathematics, neurons, chemistry, electronics, etc. That all happens each step of the way, even if the next technology is completely different.
All of which have absolutely nothing to do with what we are currently calling AI.
Doing with it, sure, but the creation of LLMs, and the algorithms behind them, especially the training, are what I’m talking about. It’s a lot of very impressive, complicated math
I think it’s pretty pathetic that “fuck AI” has become the trendy, cool thing. It really misses the mark. It should be fuck capitalism and the sociopathic CEOs abusing AI and shoving it down our throats. AI is not the problem.
It's actually just a lot of pretty simple maths from decades ago, but it's a lot of it. The big changes in those decades have been the feasibility of doing enough of that simple maths to achieve anything useful, and domain-specific network architecture stuff that's rarely transferable, e.g. LLMs are possible because of the invention of the transformer architecture in 2017, and that's also turned out to be useful for a few things like image generation and protein folding simulation, but not for all neural network based techniques, and then most of the things that have made successive LLMs better haven't also been useful for the few other transformer-architecture-based neural networks. Most not-LLM AI isn't going to be meaningfully easier to create than it would have been had the world got bored after GPT-2 and we'd only focussed on doing image and video generation.
Transformer is useful for damn near anything. At the end of the day, what we consider intelligence is the ability to predict what comes next, whether that is what our senses will tell us next or what the next hypothesis to test should be based on the data we have seen so far.
It's not damn near anything. There's loads of stuff that computers can do much more quickly and more accurately without it just by virtue of computers already being fast and effective at maths and obeying logic. With or without the transformer architecture, a neural network is never going to be as fast or reliable at, for example, summing a collection of numbers as just adding them would be, and loads of real-world tasks are like this, hence why we've built billions of computers even before the transformer architecture was invented.
Also, in particular, I didn't say that the transformer architecture wasn't useful for things that aren't LLMs, I said that most of the work done specifically to improve LLMs has no applications outside LLMs, so the next big leap towards making computers intelligent isn't helped more by working on LLMs than it would be by working on any other kind of AI.
They can use it to do a lot of things. AI is far from perfect and makes all sorts of weird mistakes, but so do people. Arguably there's substantially more value in training inexperienced humans to get better in their fields than in settling for AI as a cheap alternative that starts with a maybe slightly higher or similar but cheaper baseline, but that doesn't eliminate all value they create. You can make arguments about the long term benefits socially or for individual organizations that leverage AI, but spend a couple hours playing with Claude and it becomes extremely evident that they're not anything resembling useless.
Even if we completely throw chat bots out the window, there are some instances of general utility for thinking models. This comic is making a moral argument that's more compelling, but arguing that they're actually totally useless doesn't really reflect reality
I gotta pull you up on this one. Realistically, they're not in the same ballpark. The mistakes made by people are so minor and infrequent they can be trusted to perform brain surgery, and the kinds of mistakes made are things like 'we miscounted the sponges on 1 out of 1000 surgeries,' or 'I took too little tissue and didn't get all the red cancer out of the identically red tissue because I was trying to conserve quality of life,' while LLMs are doing the equivalent of hallucinating a spleen inside someone's head or ignoring the cancer to look for cankers. You can't even rightly call of a mistake because the LLM isn't 'trying to do something and failing.' It's just producing probabilities and we're hoping they're useful.
When it's really important to get something right, we have a person double check the work of the first person, which they can do, because they are grounded in reality. When you want to check the output of AI, you use a person for the same reason. AI has no grounding in reality, only words, which are famously not the same thing as reality if you have an intellectual age greater than seven.
If you'd like to take the next sentence into consideration and respond to the context of what I was actually saying there, feel free.
As someone who's used Claude and most other big LLMs as part of my job, they're all absolutely useless. They don't have the capacity for thought or care, all they are is a word generation algorithm similar to Cleverbot. So you can't rely on them for useful information, you can't rely on them for remembering info you told them, half the time it feels like talking to a brick wall (because you essentially are), and their only actual value is to CEOs as something they can blame layoffs on, even when it's bullshit.
What do you use them for? I use them day to day for significant automation hooked into a rag tool and several custom mcp services. They are absolutely amazing but with some serious flaws that do require significant guard rails and human in the loop points.
They are obviously over hyped and being used as an excuse by shitty CEOs but to call them useless is a mistake I think
Sorry, I’m curious: what’s your workflow looking like when you’re dealing with LLMs?
Because I‘m just tinkering with them as a hobby and while I consider them erratic and certainly limited in many regards, I still find them useful. Even fun, but on the other hand I’m not forced to use them.
I'm with you: the experiences people have with these tools are just dramatically different from mine. They are quite good. By no means even close to perfect, but they're just so much faster than me at pulling up some random information that would be hard to find with an Internet search myself and very good at going from nothing to something that works with code. I don't particularly enjoy using them because I find the whole industry abhorrent, but their usefulness isn't in question to me.
The people no longer review their broken code the catch is that if they do it would negate all the gains.
Laughable to call an LLM a thinking machine. It's glorified auto-complete built on stolen data. I work in the industry and the fact that any of this can impress anyone is fairly depressing to me.
That's not what "thinking model" means. It's not a statement about cognition, it means it takes steps in which it explains itself to itself to check if it's missing something.
Whoever coined that is using those words wrong, then.
Making a better LLM isn't the point of all this, it's taking what they have and building on it until they create a true AGI.
Whoever gets there first, makes basically everything else obsolete in an instant.
In a world where the organisations that are blazing the trail are in private hands, this is very bad news for everyone who isn't in the winning organisation.
That's essentially the arms race: who gets to be king of the world.
The slim chance of it not being monumentally detrimental to humanity is basically tied to us abandoning capitalism wholesale and uniting the world, so I'm not holding my breath.
Edit: few downvotes on this, so check my other replies for clarity, if you still think I'm taking out my arse, comment and set me right. It's Lemmy, the points don't matter, I'd rather have a conversation. Plus read again if you somehow get the impression I'm advocating for any of this
Nobody's making AGI anytime soon. LLMs do not have any of the baselines required for this. They're expensive predictive text algorithms, more or less the same ones used in mobile keyboards, but upscaled to an absurd degree. Anyone truly worried about other companies or nations developing AGI has no idea how our current "AI" works. You're never going to get there by building on them.
I'd like to believe too, but it doesn't really track when you watch what these companies are actually doing.
Of course an LLM on its own isn't going to become an AGI. Anyone with a braincell can see that. These orgs aren't so high on their own farts that they ignore this.
Nearly all of the actual uses today aren't just the LLM, but the tooling built on top of it, the LLM is the bit that you can plug into the past century of computing developments to enable much greater autonomy.
It's true to say an LLM in isolation isn't going to become AGI, but it's also looking very likely that an AGI will feature an LLM as a key component.
That's what's happening in parallel to the model development, tooling and harnesses that make the overall system more capable. If it can be done by a computer (or by extension a sufficiently advanced robot), the LLM can do it too with a bit of integration work (which it is very able to do on its own today, with minimal steering). If you can test for something being correct in any way, that too can be ultimately hooked up to an LLM as another input to push it back onto the desired path when it veers off.
Frankly I'm starting to feel like for most people it'll feel like it's years off until the day it happens. I don't see remotely enough people taking the risk seriously in time to do anything.
Wheels are a key component of my car I guess.
You know what, that's actually a very good example.
The wheels are the interface between the engine and any kind of° surface, with no prior knowledge of those surfaces
We've got an engine of basically everything that followed the industrial revolution until now.
An LLM can very much function as the wheel to marry a surface to that engine.
°Horizontal, don't be a smart ass
LOL.
In my "analogy" you just invented all of the actual complexity of the car in the same hand wavy way you claim AGI will just coalesce from the ether.
You make some interesting points.
But...the vast majority of corporate decision making for the last ten years is solid evidence that they are 100% high on their own farts.
Interesting point. But the folks giving these things autonomy are mostly just creating huge messes, right now, and then claiming victory and taking a quick bow before the stage caves in.
The places we do see success are where no human could be patient enough - which is the stuff computers were already better at, than us.
As you point out, all that can be fixed.
But it's all already not worth the money invested, before they build dozens more data centers in the hope that they can fix it. There's just massive amounts of magic thinking going on, by investors.
I do agree with your point that there's probably somethings that are useful and some that are dangerous on the other side of this.
Completely agree, but in the gap left by "vast majority", these guys don't seem to be behaving entirely like your conventional consumer squeezing companies, they seem to be playing a much more collusive game at the very least
That's the flaw in the common view of this. You're looking too short term. The second someone can offer to replace a business owners employees for half price, it's basically infinite money.
That is what they're pouring all the money in for: a chance at that prize. A chance at replacing all paid work with an automation they own.
TechBros repeat this constantly, but it just isn't true.
Plenty of second-on-the-scene solutions have emerged as most popular, or most impactful.
But Tech Bros need the fear of missing out (fear of arriving second) to justify huge investments with no worthwhile results.
I'm answering the question of what the arms race is
The goal is a technology that replaces the need for humans in any job.
The FOMO is different now because they don't give a shit about the consumer. The FOMO is versus the other competitors because it's a winner takes all scenario.
They will keep accelerating to the detriment of literally everything else.
Whether you believe they'll make it is almost moot. They're going to burn the world down trying.
That is a really good point.
True AGI is not happening within the lifetime of anyone or anything alive today.
See my other comments