• 2 Posts
  • 38 Comments
Joined 1 year ago
cake
Cake day: June 9th, 2023

help-circle
  • I said it was a neural network.

    You said it wasn’t.

    I asked you for a link.

    You told me to do your homework for you.

    I did your homework. Your homework says it’s a neural network. I suggest you read it, since I took the time to find it for you.

    Anyone who knows the first thing about neural networks knows that, yes, artificial neurons are simulated with matrix multiplications, why is why people use GPUs to do them. The simulations are not down to the molecule becuase they don’t need to be. The individual neurons are relatively simple math, but when you get into billions of something, you don’t need extreme complexity for new properties to emerge (in fact, the whole idea of emergent properties is that they arise from collections of simple things, like the rules of the Game of Life, for instance, which are far simpler than simulated neurons). Nothing about this makes me wrong about what I’m talking about for the purposes of copyright. Neural networks store concepts. They don’t archive copies of data.



  • I’m not sure why you’re asking that. You literally just asked me if I’m refusing to admit that AI could cause trouble for people’s livelihoods. I don’t know where you even got that idea. I never asked you anything about whether you admit it could help with things, because that’s irrelevant (and also it would be a pretty silly blanket assumption to make).

    Are you sure you’re not projecting here? In this entire thread, have you budged an inch based on all the people arguing against your original post?

    Who am I supposed to be budging for? Of the three people here who are actually arguing with me, you’re the only one who isn’t saying they’re going to slash my car tires and likening personal AI use to eating steak in terms of power usage (it’s not even in the same ballpark), or claiming that Stable Diffusion doesn’t use a neural network. I only replied to the other guy’s most recent comment because I don’t want to be swiftboated – people will believe other people who confidently state something that they find validating, even if they’re dead wrong.

    We just seem to mostly have a difference of opinion. I don’t get the sense that you’re making up your own facts. And fundamentally, I’m not convinced of the idea that only a small group of people deserve laws protecting their jobs from automation, particularly not at the expense of the rest of us. If we want to grant people relief from having their jobs automated away, we need to be doing that for everybody, and the answer to that isn’t copyright law.

    And as far as AI being used to automate dangerous jobs, copyright isn’t going to stop that at all. Tesla’s dangerous auto-pilot function (honestly, I have no idea if that’s a neural network or just a regular computer program) uses data that Tesla gathers themselves. Any pharmaceutical company that develops an AI for making medicines will train it on their own trade secrets. Same with AI surgeons, AI-operated heavy machinery, and so on. None of that is going to be affected by copyright, and public concerns about safety aren’t going to get in the way of stockholders and their profits anymore than it has in the past. If you want to talk about the dangers of overreliance on AI doing dangerous work, then by all means talk about that. This copyright fight, for those large companies, is a beneficial distraction.


  • You need to do your own homework. I’m not doing it for you. What I will do is lay this to rest:

    https://en.wikipedia.org/wiki/Stable_Diffusion

    Stable Diffusion is a latent diffusion model, a kind of deep generative artificial neural network. Its code and model weights have been released publicly […]

    https://jalammar.github.io/illustrated-stable-diffusion/

    The image information creator works completely in the image information space (or latent space). We’ll talk more about what that means later in the post. This property makes it faster than previous diffusion models that worked in pixel space. In technical terms, this component is made up of a UNet neural network and a scheduling algorithm.

    […]

    With this we come to see the three main components (each with its own neural network) that make up Stable Diffusion:

    • […]

    https://stable-diffusion-art.com/how-stable-diffusion-work/

    The idea of reverse diffusion is undoubtedly clever and elegant. But the million-dollar question is, “How can it be done?”

    To reverse the diffusion, we need to know how much noise is added to an image. The answer is teaching a neural network model to predict the noise added. It is called the noise predictor in Stable Diffusion. It is a U-Net model. The training goes as follows.

    […]

    It is done using a technique called the variational autoencoder. Yes, that’s precisely what the VAE files are, but I will make it crystal clear later.

    The Variational Autoencoder (VAE) neural network has two parts: (1) an encoder and (2) a decoder. The encoder compresses an image to a lower dimensional representation in the latent space. The decoder restores the image from the latent space.

    https://www.pcguide.com/apps/how-does-stable-diffusion-work/

    Stable Diffusion is a generative model that uses deep learning to create images from text. The model is based on a neural network architecture that can learn to map text descriptions to image features. This means it can create an image matching the input text description.

    https://www.vegaitglobal.com/media-center/knowledge-base/what-is-stable-diffusion-and-how-does-it-work

    Forward diffusion process is the process where more and more noise is added to the picture. Therefore, the image is taken and the noise is added in t different temporal steps where in the point T, the whole image is just the noise. Backward diffusion is a reversed process when compared to forward diffusion process where the noise from the temporal step t is iteratively removed in temporal step t-1. This process is repeated until the entire noise has been removed from the image using U-Net convolutional neural network which is, besides all of its applications in machine and deep learning, also trained to estimate the amount of noise on the image.

    So, I’ll have to give you that you’re trivially right that Stable Diffusion does use a Markov Chain, but as it turns out, I had the same misconception as you did, that that was some sort of mathematical equation. A markov chain is actually just a process where each step depends only on the step immediately before it, and it most certainly doesn’t mean that you’re right about Stable Diffusion not using a neural network. Stable Diffusion works by feeding the prompt and partly denoised image into the neural network over some given number of steps (it can do it in a single step, although the results are usually pretty messy). That in and of itself is a Markov chain. However, the piece that’s actually doing the real work (that essentially does a Rorschach test over and over) is a neural network.



  • So most of my opinions about what AI can do aren’t about hype at all, but what I’ve personally experienced with it firsthand. The news, frankly, is just as bad a source about AI is marketing departments of AI companies, because the news is primarily written by people who feel threatened by its existence and are rationalizing reasons that it’s bad, as well as amplifying bad things that they hear and, in the best case, reporting on it without really understanding what it actually does. The news is partly why you’re connecting what’s happening with that WGA/SAG-AFTRA contract; nothing I’ve said here supports people losing their existing rights to their own likenesses, and the reason they’re trying to slip it into the contracts is because even under existing copyright law, AI isn’t a get out of jail free card to produce copyrighted works despite the fact that you can train it on them.

    At any rate, here are a few of my personal experiences with using AI:

    • I’ve used AI art generation to create background art for a video game that I made with my kids over winter break, and because of that, it looks really good. It would have otherwise looked pretty bad.
    • For my online tabletop roleplaying campaign, I generate images of original locations and NPCs.
    • I subscribe to ChatGPT and because of that I have access to the GPT-4 version, which is leaps and bounds smarter than GPT-3 (although it’s still like talking to some kind of savant who knows a whole lot of information but has trouble with certain types of reasoning). While ChatGPT isn’t something you should use to write your legal briefs (I could have told you that before that dumbass lawyer even tried it), it’s an amazing replacement for google, which nowadays involves a lot of fiddling and putting quotations marks around things just so you can get a result that’s actually addressing what you want to know as opposed to “here’s a bunch of vaguely related shit that has almost nothing to do with what you asked.” That alone has improved my life.\
    • It’s also great at helping you figure out what something is called. “I’m looking for a thing that does X and Y, but I don’t know what it’s called.” Google is absolutely terrible at that.
    • I’ve used ChatGPT to generate custom one-shot adventure ideas for my online roleplaying game. Rather than having to adapt an existing adventure module to what I’m doing, if I give it information about my campaign, it’ll come up with something that utilizes my existing locations, NPCs, and setting. (Indicentally, when people say that AI “can’t be creative”, they’re essentially using a tautological definition of creativity that amount so “AI isn’t creative because only humans can be creative, therefore AI can’t be creative.” AI, in my experience, is very creative.) Compare this to the common advice that people give to game masters who can’t come up with an idea: take someone else’s story, change a few things, and drop it into your campaign. ChatGPT is also amazing at worldbuilding.

    This kind of thing is why I’m excited about AI – it’s improving my life in a big way right now. None of what I’ve done with it is “hype”. I don’t care that Elon Musk’s dumb ass is starting his own AI company, or what tech company marketing divisions have to say about it, or what some MBA CEO’s wild guess about what we’ll be using it for in 5 years is.




  • What, specifically, do you think I’m wrong about?

    If it’s the future potential of AI, that’s just a guess. AGI could be 100 years away (or financially impossible) as easily as it could be 5 years. AGI is in the future still, and nobody is really qualified to guess when it’ll come to fruition.

    If you think I’m wrong about the present potential of AI, I’ve already seen individuals with no budget use it to express themselves in ways that would have required an entire team and lots of money, and that’s where I believe its real potential right now lies. That is, opening up the possibility for regular period to express themselves in ways that were impossible for them before. If Disney starts replacing animators with AI, I’ll be right there with you boycotting them. AI should be for everyone, not for large corporations that can already afford to express themselves however they want.

    If you think I’m wrong that AIs like ChatGPT and Stable Diffusion do their computing with simulated neurons, let me know and I’ll try to find some literature about it from the source. I’ve had a lot of AI haters confidently tell me that it doesn’t (including in this thread), and I don’t know if you’re in that camp or not.



  • You will never move a boat with nuclear,

    I assume you haven’t heard of aircraft carriers and nuclear submarines.

    Also, nuclear power can be stored in batteries and capacitors and then used to move electric vehicles (including boats, planes, and tractors), so I don’t know what the hell you’re even talking about.

    Eat less meat! How hard is it to compute! So turn off your stupid AI and eat less meat. Do it now, stop eating meat.

    I’ve actually cut my meat consumption way down.

    That being said, a person using AI consumes an absolutely minuscule amount of power compared to a person eating a steak. One steak (~20kwh) is equivalent to about 60 hours of full time AI usage (300W for an nvidia A100 at max capacity), and most of the time a person spends using an AI is spent idling while they type and read, so realistically it’s a lot longer than that.

    Again, your hypothetical data center smashers are going after AI because they hate AI, not because they care about the environment. There are better targets for ecoterrorism. Like my car’s tires, internet tough guy.


  • I’m looking at it with a computer science degree and experience with AI programming libraries.

    And yes, it’s a machine that simulates neurons using math. We simulate physics with math all the way down to the quantum foam. I don’t know what your point is. Whether it’s simulated neurons or real neurons, it learns concepts, and concepts cannot be copyrighted.

    I have a sneaking suspicion since you switched tactics from googling the wrong flowchart to accusing me of not caring about workers due to a contract dispute that’s completely unrelated to anything of the copyright stuff I’m talking about, I have a feeling you at least suspect that I know what I’m talking about.

    Anyway, since you’re arguing based on personal convenience and not fact, I can’t really trust anything that you say anyway, because we’re on entirely different wavelengths. You’ve already pretty much indicated that even if I were to convince you I’m right, you’d still go on doing exactly what you’re doing, because you’re on a crusade to save a small group of your peers from automation, and damn the rest of us.

    Best of luck to you.



  • But if it makes you happy, how about we get a copyright ala Creative Commons that can allow an individual to create an AI using the copyrighted work for non-profit reason, but restrict corporations from doing so with an AI used for profit, and considers any work created by this AI to be noncopyrighted.

    Honestly, I think keeping the output of AI non-copyrighted is probably the best of both worlds, because it allows individuals to use AI as an expressive tool (you keep separating “creatives” from “average people”, which I take issue with) while making it impractical for large media companies to use.

    At any rate, the reason copyright restrictions would just kill open source AI is that it strikes me as incredibly unlikely that you’re going to be able to stop corporations from training AI on media that they own outright. Disney has a massive library of media that they can use as training data, and no amount of stopping open source AI users from training AI on copyrighted works is going to prevent Disney from doing that (same goes for Warner Bros, etc). Disney, which is known for exploiting its own workers, will almost certainly use that AI to replace their animators completely, and they’ll be within their legal rights to do so since they own all the copyrights on it.

    Now consider companies like Adobe, Artstation, and just about any other website that you can upload art to. When you sign up for those sites, you agree to their user agreement, which has standard boilerplate language that gives them a sublicenseable right to use your work however they see fit (or “for business purposes”, which means the same thing). In other words, if you’ve ever uploaded your work anywhere, you’ve already given someone else the legal right to train an AI on your work (even with a creative interpretation of copyright law that allows concepts and styles to be copyrighted), which means they’re just going to build their own AI and then sell it back to you for a monthly fee.

    But artists and writers should be compensated every time someone uses an AI trained on their work, right? Well, let’s look at ChatGPT for a moment. I have open source code out there on github, which was almost certainly included in ChatGPT’s training data. Therefore, when someone uses ChatGPT for anything (since the training data doesn’t go into a database; it just makes tiny tiny little changes to neuron connection weights), they’re using my copyrighted work, and thus they owe me a royalty. Who better to handle that royalty check but OpenAI? So now you get on there and use ChatGPT, making use of my work, and some of the “royalty fee” they’re now charging goes to me. Similarly, ChatGPT has been trained on some of whatever text you’ve added to the internet (comments, writing, whatever, it doesn’t matter), so when I use it, you get royalties. So far so good. Now OpenAI charges us both, keeps a big commission, and we both pay them $50/month for the privilege of access to all that knowledge, and we both make $20/month because people are using it, for a net -$30/month. Who wins? OpenAI. With a compensation scheme, the big corporations win every time and the rest of us lose, because it costs money to do it, and open source can’t do it at all. Better to skip the middle man, say here’s an AI that we all contributed to and we all have access to.

    So again, what specifically is your plan to slow down deployment? Because onerous copyright restrictions aren’t going to stop any of the people who need to be stopped, but they will absolutely stop the people competing with those people.


  • Lots to unpack here.

    First of all, the physical process of human inspiration is that a human looks at something, their optic nerves fire, those impulses activate other neruons in the brain, and an idea forms. That’s exactly how an AI takes “inspiration” from images. This stuff about free will and consciousness is metaphysics. There’s no meaningful difference in the actual process.

    Secondly, let’s look at this:

    SAG-AFTRA just got a contract offer that says background performers would get their likeness scanned and have it belong to the studio FOREVER so that they can simply generate these performers through AI.

    This is what is happening RIGHT NOW. And you want to compare the output of an AI to a human’s blood sweat and tears, and argue that copyright protections would HURT people rather than help them avoid exploitation.

    I’ll say right off that I don’t appreciate the “you’re a bad person” schtick. Switching to personal attacks stinks of desperation. Plus, your personal attack on me isn’t even correct, because I don’t approve of the situation you described any more than you do. The reason they’re trying to slip that into those people’s contracts is because those people own their likenesses under existing copyright law. That is, you don’t have to come up with a funny interpretation of copyright law where concepts can be copyrighted but only if a machine learns them. They need a license to use those people’s likenesses regardless of whether they use an AI or Photoshop or just have a painter do it. Using AI doesn’t get them out of that – if it did; they wouldn’t need to try to put it into the contract.

    In other words, they aren’t using an AI to attack anyone; they’re using a powerful bargaining position to try to get people to sign away an established right they already have according to copyright law. That has absolutely nothing to do with anything I’m talking about here, except that you want to attach it to what I’m talking about so you can have something to rage about.

    And here’s the thing. None of you people ever gave a shit when anybody else’s job was automated away. Cashiers have had their work automated away recently and all I hear is “ThAt’S oKaY bEcAuSe tHeIr jOb sUcKs!!!111” Artists have been actually violating the real copyright of other artists (NOT JUST LEARNING CONCEPTS) with fanart (which is a DERIVATIVE WORK OF A COPYRIGHTED CHARACTER) for god only knows how long and there’s certainly never been a big outcry about that.

    It sucks to be the ones looking down the business end of automation. I know that because as a computer programmer I am too. On the other hand, I can see past the end of my own nose, and I know how amazing it would be if lots of regular people suddenly had the ability to do the things that I do, so I’m not going to sit there and creatively interpret copyright law in an attempt to prevent that from happening. If you’re worried about the effects of automation, you need to start thinking about things like a universal healthcare and universal income, not just ESTABLISH SPECIAL PROTECTIONS FOR A TINY SUBSET OF PEOPLE WHOM YOU HAPPEN TO LIKE. It just seems a bit convenient, and (dare I say) selfish that the point in history that we need to start smashing the machines happens to be right now. Why not the printing press or the cotton gin or machines that build railroads or looms or or robots in factories or grocery store kiosks? The transition sucked for all those people as well. It’s going to suck for artists, and it’ll suck for me, but in the end we can pull through and be better off for it, rather than killing the technology in its infancy and calling everyone a monster who doesn’t believe that you and you alone ought to have special privileges.

    We need to be using the political clout we have to push us toward a workable post-scarcity economy, as opposed to trying to preserve a single, tiny bit of scarcity so a small group of people can continue to do something while everybody else is automated away and we all end up ruled by a bunch of rent-seeking corporations. Your gatekeeping of the ability of people to do art isn’t going to prevent any of that.

    P.S. We seem to be at the very beginning of a major climate disaster these last couple weeks, so we’re probably all equally fucked anyway.





  • If what you’re going to give me is an oversimplified analogy that puts too much faith in what AI devs are trying to sell and not enough faith in what a human brain is doing, then don’t bother because I will dismiss it as a fairy tale.

    I’m curious, how do you feel about global warming? Do you pick and choose the scientists you listen to? You know that the people who develop these AIs are computer scientists and researchers, right?

    If you’re a global warming denier, at least you’re consistent. But if out of one side of you’re mouth you’re calling what AI researchers talk about a “fairy tail”, and out of the other side of your mouth you’re criticizing other people for ignoring science when it suits them, then maybe you need to take time for introspection.

    You can stop reading here. The rest of this is for people who are actually curious, and you’ve clearly made up your mind. Until you’ve actually learned a bit about how they actually work, though, you have absolutely no business opining about how policies ought to apply to them, because your views are rooted in misconceptions.

    In any case, curious folks, I’m sure there are fancy flowcharts around about how data flows through the human brain as well. The human brain is arranged in groups of neurons that feed back into each other, where as an AI neural network is arranged in more ordered layers. There structure isn’t precisely the same. Notably, an AI (at least, as they are commonly structured right now) doesn’t experience “time” per se, because once it’s been trained its neural connections don’t change anymore. As it turns out, consciousness isn’t necessary for learning and reasoning as the parent comment seems to think.

    Human brains and neural networks are similar in the way that I explained in my original comment – neither of them store a database, neither of them do statistical analysis or take averages, and both learn concepts by making modifications to their neural connections (a human does this all the time, whereas an AI does this only while it’s being trained). The actual neural network in the above diagram that OP googled and pasted in here lives in the “feed forward” boxes. That’s where the actual reasoning and learning is being done. As this particular diagram is a diagram of the entire system and not a diagram of the layers of the feed-forward network, it’s not even the right diagram to be comparing to the human brain (although again, the structures wouldn’t match up exactly).