AIs gunning for our precious freelancers

[rotting bones]

I can’t (be bothered to learn how to) cut and paste on the iPad, but the discussion of E=mc^2 struck me as not understanding how new science works. It’s not a matter of rearranging letters and deciding on the exponents.

Like I said, I agree that AIs are of limited utility in higher thought. It's not just in math or science, but also philosophy.

Einstein famously started by asking what the world would look like if you could travel at the speed of light. An LLM can’t start from that and derive special relativity, nor could most humans. With the caveat that *now that many explanations of relativity exist*, it can generate a pastiche of such an explanation.

Actually, there are theorem proving LLMs that can help you draw out the consequences of certain theories, but that's another issue. I was only talking about:

1. How AIs learn functions. These can derive approximations to E=mc^2 from data regardless of what thought experiment you can use to argue for it it. Language is not the only thing neural networks can predict.

2. How the ChatGPT tries to master English idiom.

Nor is there any need to develop artificial Einsteins.

I used to be a big AI fan, back when it was a fascinating programming problem. Now that we know how it would be used— to enrich capitalists and emmiserate actual humans— I think GAI would be a mistake. Maybe a socialist utopia could safely make one—as a valued co-worker, not a slave— but that can wait till we have the utopia.

All improvements under capitalism are weaponized against some group of workers. This line of thinking leads to accelerationism.

It's a chicken and egg problem. 1. I don't think people will go for socialism if the technology to support it doesn't exist. 2. If the technology for socialism is created, that will make life worse for many people under capitalism.

All we can do is both create the technology and fight for socialism at the same time.

Also, we can't not have AIs at this point. You can't put knowledge back where it came from. My suggestion was to make using AI art illegal for commercial applications, since that's what we can control.

My brother is an artist who volunteers artwork and patronizes other artists using his day job. He uses ChatGPT too these days. Sometimes he only uses it to visualize arrangements of scenes before drawing them himself. Artists with big studios hire assistants too, etc.

[Travis B.]

Case in point, at my work we use 'deep learning' for MR image processing, and it does not compare to generative 'AI' in any outward respect. If you did not know the internals of our software you could very well have not a clue that 'AI' techniques are being used to generate MR images.

Redirecting the topic for a moment: might you be able to give me a comprehensible explanation of what ‘deep learning’ is supposed to be? I’ve been hearing the term for a long time, but I’ve yet to work out what it means beyond ‘a neural network with lots of hidden layers’. In which case LLMs should qualify under that name too…

Frankly, I don't really deal with the 'AI' side of things, but I gather it probably means what you think it means ─ 'a neural network with lots of hidden layers'. What I meant is our use of what we call 'deep learning' does not really resemble prompt-driven LLM's and image generators, and rather is a hidden image processing layer used to de-noise and otherwise clean up image data, such that if you did not already knew we used it you would never guess that 'AI' was used in generating our images.

[rotting bones]

Frankly, I don't really deal with the 'AI' side of things, but I gather it probably means what you think it means ─ 'a neural network with lots of hidden layers'. What I meant is our use of what we call 'deep learning' does not really resemble prompt-driven LLM's and image generators, and rather is a hidden image processing layer used to de-noise and otherwise clean up image data, such that if you did not already knew we used it you would never guess that 'AI' was used in generating our images.

"Deep learning" is a marketing buzzword. You're thinking of classification as opposed to regression, clustering, etc.

Classifiers can be much more accurate than generative models.

[zompist]

Rotting: I agree that LLMs are here to stay. I don’t see a need to develop GAI.

Of course, we don’t know how far away GAI is. If you ask a techbro, it’s next quarter. I suspect it will require another paradigm shift— could be decades.

Also, are you under the impression that special relativity was a matter of fitting equations to data? There was no data. All the supporting data came *after* the theory.

[Richard W]

Also, are you under the impression that special relativity was a matter of fitting equations to data? There was no data. All the supporting data came *after* the theory.

The Michelson-Morley experiment and the Fitzgerald-Lorentz contraction came before. What you say is true of general relativity.

[zompist]

Also, are you under the impression that special relativity was a matter of fitting equations to data? There was no data. All the supporting data came *after* the theory.

The Michelson-Morley experiment and the Fitzgerald-Lorentz contraction came before. What you say is true of general relativity.

That’s not “data”, that’s one datum (MM’s null result) and a theory.

Whether another theory could have worked I don’t know, but compare the problem of fitting a line to a set of points: infinitely many lines fit a single point.

[Raphael]

Well, to get really pedantic about it, Michelson-Morley only proved that c is a constant from the perspective of the light source. If we would live in a universe where the speed of any given light wave is always c plus the speed of the light source, Michelson-Morley would have had the same result. That c is constant from everyone's perspective was, IIRC, only demonstrated by astronomical observations a few years after 1905.

IIRC, Einstein's starting point was simply that Special Relativity was the only way to resolve an apparent contradiction between classical mechanics as it was understood before Special Relativity, and classical electrodynamics, again, as it was understood before Special Relativity.

Of course, none of that was about being very good at doing arithmetic in your head.

[Richard W]

Michelson-Morley is a set of data points, unless you’re willing to accept that the ether happens.to move with the Earth.

[bradrn]

Michelson-Morley is a set of data points, unless you’re willing to accept that the ether happens.to move with the Earth.

And, IIRC, plenty of scientists at the time were willing to accept just that.

[Raphael]

My point was more like, in 1905, even if you were willing to dismiss ether, "c is always constant" didn't instantly follow.

[zompist]

Really interesting development in AI: a Chinese company called DeepSeek has released an extremely lightweight LLM that at least matches OpenAI's o1, at about 3% of the cost. It uses lower-cost Intel chips to get around US export restrictions.

https://www.nature.com/articles/d41586-025-00229-6

https://www.wired.com/story/deepseek-china-model-ai/

This is a big kick in the pants to US-based AI, which has sunk billions into its LLMs. Currently OpenAI is losing about $5 billion a year. DeepSeek comes out of a Chinese hedge fund worth $8 billion, but not all of that went into the product; some reports say its LLM cost millions. It looks like US protectionism backfired spectacularly, spurring innovation rather than shutting it down.

They're also open-sourcing the model itself, though not the training data. That's got to send a shiver through the spines of the techbros. Their whole idea was to keep AI huge and costly with themselves as the profit-makers. If AI is instead cheap and universal, it's not a profit center for anyone.

Now, I'm no fan of LLMs, but one reason for that is their enormous cost and the domination of tech scumbags. And the other major reason is the devaluation of human creativity; but maybe that is lessened if the techbros can't replace payments to artists with payments to themselves.

[malloc]

They're also open-sourcing the model itself, though not the training data. That's got to send a shiver through the spines of the techbros. Their whole idea was to keep AI huge and costly with themselves as the profit-makers. If AI is instead cheap and universal, it's not a profit center for anyone.

Sure, but the greater availability of generative AI means that humans face even more competition than ever. The question that none of you techies has ever been able to answer is what becomes of humans when AI and other forms of automation become powerful enough to replace us at all or most jobs. Much as scientific progress has created a "God in the gaps" problem, technological progress is steadily narrowing the gap into which humanity can fit.

[bradrn]

you techies

That’s a bit unfair, I think. Many of us are programmers but not all, and I’m not sure anyone here subscribes to the stereotypical ‘techie’ views.

(Now, I don’t have an answer to the question either. But I don’t work in programming, and in fact go out of my way to avoid AI. There’s also too much uncertainty in the actual effects of AI for me to feel comfortable hazarding a guess.)

[malloc]

That’s a bit unfair, I think. Many of us are programmers but not all, and I’m not sure anyone here subscribes to the stereotypical ‘techie’ views.

Sorry if that sounds harsh but you must understand my intense anger and frustration at this issue. The tech industry wields unprecedented power over the world, as evidenced both by radical technology like AI and the fantastical wealth of its leaders. It has abused that power to kick the human race while it's already down, struggling for its very existence against both fascism and global warming. Since time immemorial, art and literature have been refuges from the horrors of the world. Faced with implacable dictatorship and atrocities, people used art and literature to make sense of their experiences. Now the tech industry wants to eliminate the artist and writer. Furthermore the tech industry has itself become the beating black heart of the reactionary wave sweeping the world right now.

[zompist]

They're also open-sourcing the model itself, though not the training data. That's got to send a shiver through the spines of the techbros. Their whole idea was to keep AI huge and costly with themselves as the profit-makers. If AI is instead cheap and universal, it's not a profit center for anyone.

Sure, but the greater availability of generative AI means that humans face even more competition than ever. The question that none of you techies has ever been able to answer is what becomes of humans when AI and other forms of automation become powerful enough to replace us at all or most jobs.

It's been answered many times. What's strange is that you don't accept the answer even though it should be in line with your worldview: the problem is not AI per se, it's capitalist techbros. Altman et al. want to create new quasi-monopolies that cost and reap billions of dollars, forever. And as I just said, they want to get the money paid to artists to go to them instead.

DeepSeek is very bad news for them. If AI is cheap (3% of OpenAI's price) and doesn't require ecosphere-destroying levels of energy, then that attempted capitalist takeover becomes impossible.

That does leave the problem of CEOs who think that LLMs can replace their workers. Ironically, cheap AI may not be as exciting to CEOs. CEOs get excited by huge technology; perhaps less so if it's more like buying a computer. If the AI hype machine dies, so much the better.

The fact that DeepSeek is open-source is also extremely interesting. Is it better for artists? I have no idea, but it does mean that the artists, and not just the techbros, have access to cheap tools. Democratizing cheap innovations generally works better for people than multi-billion-dollar monopolies.

I'd also remind you that-- as I've been saying since this thread started-- you've been swallowing the techbro propaganda that GAI is just around the corner. There are real things to worry about, and it's precisely to their advantage to get you to worry about GAI instead.

[malloc]

It's been answered many times. What's strange is that you don't accept the answer even though it should be in line with your worldview: the problem is not AI per se, it's capitalist techbros. Altman et al. want to create new quasi-monopolies that cost and reap billions of dollars, forever. And as I just said, they want to get the money paid to artists to go to them instead.

But there simply is no getting rid of capitalism for the foreseeable future, regardless of my feelings toward it. Imagining a world where people like Elon Musk have suddenly developed the yips regarding propaganda and maintaining power makes little practical sense. We must accept the premise that they have the power of life and death over us as a given and oppose anything that will lead them to deem us "life unworthy of life". Considering that Musk has already revealed his true colors, I consider that phrase quite literal. We are facing a horrific situation all around but the last thing we need is for the bottom to drop out completely.

The fact that DeepSeek is open-source is also extremely interesting. Is it better for artists? I have no idea, but it does mean that the artists, and not just the techbros, have access to cheap tools. Democratizing cheap innovations generally works better for people than multi-billion-dollar monopolies.

It certainly isn't better for artists because it still leaves everyone else with no reason to hire them just like existing AI models. Once everyone has access to completely free software that can draw anything they want instantly, they have no practical reason to hire artists.

I'd also remind you that-- as I've been saying since this thread started-- you've been swallowing the techbro propaganda that GAI is just around the corner. There are real things to worry about, and it's precisely to their advantage to get you to worry about GAI instead.

It really is difficult to gauge whether the tech industry is bluffing on this point or whether AGI is looming on the horizon. Unless everything we know about neuroscience is wrong, there is nothing that prevents computers from mastering all the cognitive abilities of humans. Considering the extraordinary wealth of the tech industry and its obsession with achieving AGI, it feels careless to dismiss it as mere vaporware based on the slight possibility that materialist neuroscience is wrong.

[Ketsuban]

Zomp called you out in the post you quoted for this kind of argument, where you jump straight from "there is no universal law prohibiting X" to "X is going to happen" purely because someone with a lot of money wants X. LLMs and latent diffusion models are not AGI, and the fact people have quickly found their limitations (they know nothing[1], they have no concept of truth, they rely on the availability of massive quantities of work already done by humans to function, and cannot output anything that is not at least somewhat represented in the training data[2]) speaks to their incapacity to replace human artists

[1] which is why it's so frustrating that people try to use them as search engines, only to get nothing helpful and post their question on Reddit along with the ChatGPT output as if that's any more use than the TempleOS "digital oracle" or Zomp's Markov chain trained on Lovecraft and Lewis Carroll
[2] hence my skepticism of "ethical models" being useful for anything whatsoever—I've seen the copyright-free material proponents are suggesting would be used as their training data, and I don't think anyone has any use for a model for generating fake 19th-century political cartoons and newspaper adverts

[alice]

In any case, won't there be Men With Guns to protect us from rogue AI's which have escaped the control of their creators?

[zompist]

The tech market is terrified of DeepSeek:

* NVidia lost $600 million in market value, or 17%. In one day it went from the highest to the 3rd-highest-valued stock.
* Meta and Google were also down.
* Energy companies, which had planned to profit from the immense energy needs of US AI, plunged 21%.
* The tech-heavy Nasdaq dropped by 3.1%.
* Maybe ironically, the Dow was up 0.7%. This was seen as pressure coming off health and other services companies which were threatened by AI. (But take analysts' expalanations of stock price changes with a truckload of salt.)

[malloc]

Zomp called you out in the post you quoted for this kind of argument, where you jump straight from "there is no universal law prohibiting X" to "X is going to happen" purely because someone with a lot of money wants X. LLMs and latent diffusion models are not AGI, and the fact people have quickly found their limitations (they know nothing[1], they have no concept of truth, they rely on the availability of massive quantities of work already done by humans to function, and cannot output anything that is not at least somewhat represented in the training data[2]) speaks to their incapacity to replace human artists

But logically if something is physically possible and has trillions of dollars dedicated to its realization, it will eventually happen. Zompist himself already admitted that they were half-way to AGI and that was several years ago when the technology first debuted. Sure the current forms of AI fall short of human intelligence, but we are talking about technology that has only existed for several years. Airplanes in their first few years could barely get off the ground but eventually even the crappiest plane could outfly even the fastest and most agile bird.

Meanwhile the internet is overflowing with AI generated images and artists are struggling to find work because AI can replicate their abilities except faster and more cheaply. Perhaps the finest and most innovative artists have nothing to fear currently, but those just getting started or lacking superlative talent cannot easily keep pace with machines that can produce decent if not necessarily inspired images.