I see his point that the very best of the best humans will still continue to create for a long time. However, I see a mass homogenization of all popular media in the future. Similar to how 99% of Unity/Unreal games look the same, if you provide people with an out-of-the-box easy solution, they won't bother to come up with their own unique solution. This is something that Blow talks about a lot with using frameworks. Also, every master has to go through a journeyman phase. AI replacing the journeyman could potentially prevent a lot of human potential in the future. I don't have to worry about that thankfully but worry about future generations.
The problem is that right now you'd have to review the AI generated code to make sure it's not complete nonsense, which means you did nothing other than changing your work from writing code to reviewing code, which is more tedious. Not to mention in many cases knowing what the code is supposed to do is the actual work, so those AI systems right now are just a better intellisense in the best case scenario. Which is fine but it's not what many hyped people expect.
Completely correct. There is also the problem of application and things being adjusted for "software as a service." I have objections to SAAS but it's beside the point. I think general articial intelligence is required in order to actually run such a product because it would need a mind for the problem space. Artwork is actually REALLY flexible when it comes to satisfying a need or application, software is a bit different.
The problem is not wether the models scale forever, but of their fast adoption rate in the socioeconomic sphere. If generative LLMs become a baseline dependency anywhere close to the internet in most markets, then I'm afraid we'll generate new sensibilities which would be "okay" with subpar output. And, because it is so much cheaper and faster, we might devalue that combo of effort+creativity to the point of it becoming not worth for anyone to pursue, as it would always hit a brick wall of endless saturation and expectations too quick for any direct human input. It becomes a race to the bottom... The middle class is fucked.
I don't know, I feel like similar change have happened in different industries before. Like the industrial revolution, and producing as scale for cheaper. Like IKEA can be seens as producing an "okay" subpar output compared to artisanal woodcraft. All in all it doesn't seem so bad. cheap product have some place in the market. IMO a lot of place where illustration was just not affordable will now be able to exist.
@@ZoobTheElement huge societal upheaval needed to happen during the Industrial Revolution so that people weren't just left to die. You live in the here and the now, not in 20 years in the future where you can enjoy the positive outcomes of this tech's implementation without suffering its immediate direct results on a socioeconomic sphere not ready for it. You too have a stake in this. Also there shall be so much of artificial "content" that, unlike something made in the material world, shall completely blur out any modicum of reality out of the picture. Also, art, or creativity in general, does not have a direct assigned functionality in order for a bare minimum to be met and mass produced forever, it is manifest in the process itself and molded by the wills of those making it. Creativity is one of the most important aspects of our thinking and it should not be outsource en mass. You can't have a whole society of individuals who dont know how to do anything yet are entitled to feel themselves the creators of endless products which they know nothing off of the process behind each of their supposed fields.
This makes the assumption that the average product produced by humans is better than what can be produced by a generative LLMs. I can't speak to other domains, but when it comes to coding common patterns, ChatGPT 4.0 often produces results that have better structure than the average developer -- assuming you understand the context to use the outputs. This is because these models have weighting and selected training to choose sources that are better than what you might find if you sampled a random developer. Are there examples of bad output? Sure, and for anecdotes that you can find about bad AI you can find analogous examples in people; something that people tend to discredit when discussing generative models. Now, as Jon said, above average talent will be in demand for a long time. Having the ability to think, utilize context, and perform better the norm will be valuable as long as AGI doesn't exist. Is the middle class fucked? Probably within the near future, and that's largely a function of AI being able to replicate common patterns and work flows, which constitute the majority of middle class work. Should we as a society figure out how to support our community as we find ways to replace their jobs? Absolutely, and there will likely be a lot of growing pains with this shift in society.
Very good take from JB once again. I don't consider myself a good programmer, but I can't imagine myself using AI to help with anything beyond small and/or tedious tasks. If I were to order it to make a whole program, whatever it may be, I wouldn't be able to trust to not fuck something(s) up somewhere, and debugging that would not be a pleasant experience. This is by no means a diss at modern LLMs, as the fact that they are capable of doing even that just from a prompt in the first place is truly a miracles of science .
i think AI should not be confused with a person in the respect that it is expected to do all by itself but i also think AI can perfectly make a whole program and only be tested/judged by a real person before it enters real time production. AI should be the mean and not the end by itself. humans are the end
@@gabrielchuede6688 "and only be tested/judged by a real person" runs heavily into "and debugging that would not be a pleasant experience". Any experienced programmer will tell you that understanding others' code (and possibly refactoring it) can be times harder than writing your own code. (And let's not even go into stuff like intentional flaws in the code which can be missed by anyone not "very" proficient in their field - I've seen stuff like encryption function that looks okay and works okay... at first, once you encrypt larger thing, it fails to encrypt, and you might not even notice. And the change compared to the normal one is tiny, one line that looks like it should work, but it doesn't, and you need great understanding of bitwise operations and encryption process to gather why it fails to begin with. Now let's imagine your LLM goes through the internet in search of such encryption function for your program and implements this one instead of the main one. Why? Nobody knows, it's LLM!)
And the small tedious tasks are usually a decent break from those that require harder thinking. Or so ingrained that you can do it quickly with smart use of tools and just raw muscle memory, making it effectively nothing but a neat novelty, but nothing impressive.
@@samuelbucher5189 well, I'm happy that my overarching point got across well enough, that the only contention was whether "getting less hard work done between hard work" or "taking a total break" is better.
Also a human artist doesn’t necessarily train himself on other artists’ work. They can see something in the real world and then in their minds imagine a way to paint that, all without needing to train on someone else’s work
My wife is a digital artist. She still has more commission work than she can handle. AI just isn't on her level and can't do the stuff she does (AI doesn't understand layers, which are necessary for some work such as high-level vtuber models that have 100+ animated layers). It does, however, help with references. Think of how Pinterest works. That is what can be automated nicely for artists. Tons of references which are good images, but not consistent with what you asked (and full of small problems that would prevent them from being usable as-is).
Isn't it essential for a reference to be accurate? If I wanted to draw one, why would I look at an AI tiger where the perspective is just completely messed up and the anatomy weirdly shifted?
@@hexcodeff6624 that's the thing, not at all. For example, some newer AIs have better prompt following and consistency than the SD 1.5 based one she likes, even if the image is 512x768 she doesn't care, she likes those ones. No matter if the hands are messed up, or if the background doesn't make too much sense; she doesn't trace the image or anything - as long as some details look cool it gets her creative juices flowing. She has no problem drawing perfect hands by herself, so AI is more of a "throw me some ideas".
@@hexcodeff6624 her "actual" references also aren't like the final product. Maybe she liked the pattern on a plaid or the way light was hitting a character. She uses 3D stuff like DesignDoll or CSP's for anatomy.
@@4.0.4 Yeah, I just don't think reference is the right word for the AI stuff here. Inspiration, maybe. It is, after all, not much more than a glorified Image search algorithm.
As a artist I kind of agree about the notion of work, but at the same time its hard to not notice the stark contrast between. Oh AI wont be able to program in a long time while also saying, I did not expect to see AI art to be so good in my own lifetime.
AI has an "outrunning the bear" problem. It doesn't need to be a general intelligence, it just has to convince enough humans that it's a general intelligence.
I only disagree with his stance on how far we are with coding. Yeah, it's nowhere near perfect, but there are still so many mediocre programmers who make everything they work on worse, at which point having it done by an AI that explicitly broadcasts what it's doing is just... more sensible. Not that it matters, because all that changes in a matter of months anyway. But still, most professions have shortcomings we could just abstract away.
@henriquemarques6196 If there are no jobs then you won't need a job... How is this a problem? We do jobs in order to produce the things that we as humans need and want, so if there are no jobs, we will have the things we need and want produced by AI.
AI art has an abject banal mad magazine aesthetic to it. That's the problem they have a latent quality common to everything they output and it's subtly off-putting. If people saturate media with those images it's not good. Using chat gpt to quickly look up documentation though is useful.
I google "Victorian Architecture" and most of the first image results where Ai generated. I could tell, dunno if the average person could. Unchecked, this could become really devious. A lot of articles I've had the feeling that they ar e AI generated, but I can't tell.
@@BinaryDoodhopefully this continues and the internet is drawn to its conclusion as a medium too open to robots for its own good. Bring back books, catalogues, human curation etc.
@@krunkle5136 The lack of curation and editorship is one of the main issues with the internet, it can't be done by a robot. Glad that other people see it too.
@@krunkle5136 It's people using AI to circumvent IP ownership, not robots. People will sell books and catalogues and what not generated by AI for the same reasons they post it online.
we need hundreds of thousands of jobs for bad artists for the field to be strong enough to profuce a few actually good outliers. Good artists are a product of a system, if the skill ceiling gets raised so high you will only get hired if you're one of the top specialiats in your field, it's doomed.
Not hypothetical, I've used AI to code. It helped me get started in a project. I ended up rewriting everything it did and taking a different approach. I would have been probably faster without it but I was stuck, so it did help. People are way too scared. It's just a tool. A bit eerie, but the sky isn't falling.
similar experience, using chatgpt4 definitely helps me start and unblock me in some projects, even though I end up rewriting most if not everything, it's also great for getting some feedback on what some obfuscated code is doing if I'm not seeing it immediately
it don't work on lowlevel stuffs actually its terrible if you try to do something that has not been done before or is not that present on the web they start hallucinating . To express my taught i think for now the language models are wrong they are not designed to work with hard problems and especially novel ones, they have to come up with other models that are not manly driven by knowledge more about generating new patterns from existing ones even though they do that in some way it's still too slow and inefficient still requiring a generalization of the pattern you aim for . I would also like to add to my previous statements that even in high level tasks they still hallucinate if you go out of the norms and trends thus you have to take it with a grain of salt . That being said they still have their uses as you just pointed it out it's great to get you started or search a general information i think they do it better than a search engine in certain cases .
Depends on what you mean by theft. If you publish something without saying anything, it can legitimally be used to train a real artist and a neural network. If you explicitly forbid it from being fed to machines (and most art wasn't), then if you agree with intellectual property rights then yeah by all means just ban that stuff, but that won't stop the changes.
Jobs are infinite. As a matter of fact, "jobs" is a term used by the status quo to denote what the financial and political sector of a society wants to be done, having a general idea of who can perform what based on past experience. If a politician or banker decides that making holes in his garden is a worthy endeveour for his amusement, he will hire a company expert in hole diging and they themselves will hire low skilled workers to shovel soil out of the garden if needed (for customized hole shapes a generic driller can't perform).
I think it's one of those things where it's just up to artists to gatekeep AI from their space it if they wanna gatekeep. The main thing I see potentially becoming more of a thing with AI art is more of low skilled artists using AI to generate images and then just manipulate them in photoshop to sorta cheat their way to somewhat decent looking art quick that they otherwise wouldn't really be able to draw at their skill level. I think stuff like that should be faced with a ton of gatekeeping from the art community, and I think it generally has been.
The problem isn't artists losing their jobs, the problem is why we still need jobs to survive in an increasingly automated society. We should be excited about AI replacing human labor; only in capitalism is the loss of jobs a death sentence for everyone who gets replaced. We can't keep "making new jobs" forever once AGI and advanced robotics are on earth. Humans will simply be unemployable in every industry. Under capitalism, that means 99% of the population starving to death. Think what you will about capitalism, but it is quickly becoming outdated.
@@BinaryDood Something like Global Basic Income. If the AI and automation is generating wealth for very little cost, the price of items should be a lot cheaper, meaning GBI needn't be prohibitively expensive for society. See countries with good social welfare already.
People need to involve themselves in activities that benefit their community and require skill and effort or else people atrophy. People need work, ideally work where they're not mistreated (much of post industrial service jobs are mistreatment of the soul). Anyone saying otherwise should check with someone to make sure they don't have major depression.
@@krunkle5136 that's very humane of you! I agree pretty much with everything you said. I do like tech, but sometimes most of its possible permeations are highly disrupting and chaotic.
People are going to have to realize that, even outside of AI, many of the job markets will be unreliable in the future. Technology changes the jobs available, and that has always been true. The problem is that we've tied people's ability to consume too tightly to their ability to produce something unique, so when a machine can produce what a person can, suddenly the person has lost the ability to consume. You COULD control how people make art, but then you're creating a convoluted system to artificially keep employment up. You're also limiting the creative output of what a writer can make with a computer, which might be violating the creative rights of the writer. Perhaps the solution to these employment problems will eventually be the government paying for continuing education for adults that suddenly need to switch jobs. Of course people may also need a UBI if they have a family and need to go back to school. The whole thing is very complicated. But again, it's probably foolish to try to hold back technology for the purpose of keeping jobs. You are more likely to pass legislation for a stronger social safety net than you are to pass legislation limiting what a computer can observe. And yes, people need more empathy for anyone feeling that their jobs could be replaced by a machine, but "is it good that this happens" is a separate discussion from "will this happen".
Well, here's the thing: he's a professional programmer, and Jonathan correctly states that AI cannot solve programing. He does this because he is knowledgeable about the topic. But Jon is not an artist, yet he confidently states that we shouldn't listen to Artists, then proceeds with a weird philosopher king analogy involving horse trainers. He then callously states that he doesn't care if Artists (or whatever he refers to as "low skill" and "mediocre") don't have jobs. And as before, in a topic he knows nothing about, he again proceeds with another weird analogy, this time about coal miners. It is fine to want to have an authoritative opinion on most things, but the problem is that he painfully doesn't know about what he is talking about in some topics, yet he proceeds with an uninformed opinion anyways. Even though all things considered, his first answer applies pretty well to both Art and programming.
Artists arent horses. Art isnt about leading you from point A to point B. If you continue with that analogy then you are also willing for a lot of people to die, as that was the case for horses. It's weird. One of the main themes of The Witness to me was precisely that. The obsession of the process itself, translating the surrounding world with abstract thought to then draw back the line onto it. It really makes me sad. Seems like its not the way he thought the "collapse of civilization" would happen, though it absolutely fits the broader vision he implied (because yes, not just artists are on this tech's line of sight).
This really isn't an uninformed take. He believes: - People becoming redundant due to new technology is inevitable and fine, you just need to give them enough time to reskill - People already learn art by accessing free resources en masse and mimicking it until you can't recognise their exact sources, generative AI does this at scale with the only difference being that it's a bot not a human You can disagree, but these are reasonable and informed stances.
@@MrLordFireDragon there are places where redundancy should never be applied. Like outsourcing all creativity. It is in thinking itself, not just "art as job". Anyone who can see more than a palm in front of them and isnt convenientely narrowviewing the issue, would know the chaotic ramifications of this. Also, huge societal upheaval is often downright needed to even out such level of displacement, so countless (literally experts here) arent left to die by being cast aside whilst the very thing they made is used aganist them infinitely, with no say on the mather. And that is untrue. Read Harold Speed to realize that is an insuficient euristic. Humans must first breakdown their own utilitary symbolic interpretations of the world which the mind decodes from other senses that arent vision. Unlearning the method of signs to learn the method of seeing. The organic process of seeing the natural world from reference, or other art, is not an input -(blackbox)- output. Much less of countless finished readymade tobe consumed outputs. It requires your senses, your conscious learning, physicality of the process, recreation as opposed to merging (white canvas as opposed to noise) and a stake in life altogether, which guides your actions. You are thinking that the map comes before the place.
@@MrLordFireDragon the idea that what ai is doing is akin to what a human being is doing when viewing art is complete horseshit. You probably have next to no experience with making art just like Jonathan.
Not caring that Jr. Artists will be replaced sort of goes against his whole "collapse of civilization" narrative. Art is a stronghold of culture and therefore civilizations. To blindly give it up to automation seems fatalistic. Love the guy but as an artist, this one caught me in the fee fees 😢
1. If junior artists being replaced is furthering the collapse of civilization, then it would be supporting his idea of collapse of civilization 2. the copyright/capitalist model has been massively destructive to society (not only artists), and things like this are just exposing a flaw that already existed. I think this stuff needs to burn itself to the ground before something better can grow from the ashes, this is just how the cycle of civilizations goes. 3. artists actually have a ton of opportunity now, relatively speaking. Indeed to continue with the "collapse of civilization" theme, the collapse is when opportunity really opens up. AI spitting out these tacky images (that will become tackier as more people get used to AI art and it is classified as "fake" and "lame") is not a real threat to anyone with ambition. Those with artistic sense will see AI as another tool in their kit.
@@chrisc7265 I agree with most of what you said. What I've been saying is the "hand-drawn" artist will be more valuable in the coming years than AI models. Similar to a luthier who makes guitars by hand; you can go to Guitar Center and pick up a manufactured product for $200 or pay someone 9K to do it by hand. The human element will always be valuable.
I've noticed that when interacting with "AI stuff" Jon Blow apparently ceases to believe his "collapse of civilization" talk and most of his takes are conflicting and were addressed back then. (the talk even have an AI related question, I've checked). I find this phenomena very weird.
Industry "artists" are not artists anyway, they are industry workers, which is why their jobs will be automated. They are a cog in a machine waiting to be changed to a fancier and better cog. What I would call actual artists serve a completely different purpose and they will be fine either with or without AI.
Current AI models have no way of getting experience like humans do, so their "intellegence", if it is ever spawned, would be quirky and not reason nor flake at reasoning like humans.
There's a lot of misconceptions in this video. For example, language models are counterintuitive, because they don't actually use language. What they are is text prediction algorithms. For question answering, they just process text from websites like reddit. Because the they're trained on question answering, they are able to answer questions, by predicting the answer.
You can use more complex methods or "agentic workflows" to break down and process complicated problems in the same way. Unlike basic chatbots, these make use of reasoning loops (thinking out loud, in text form) and simplified outputs (like JSON objects/methods)
You seem to be the one with misconceptions. In order to get good at predicting text, they DID have to learn some language stuff. It's naive to say they don't use language. They use it (even if not perfectly) to predict text. They don't process text for question answering. They don't have "real" text stored anythwere, they just have a neural network. It has text in the same way that you have images in your memory.
@@neopabo No you can't. LLMs (or neural networks in general) are capable of producing useful answers in a way that was not possible before. No better way to do these things has been shown, which doesn't use neural networks.
The code being output for huge programs is vapid nonsense, but code for individual modules is on point I find that being able to spend more time architechting and specifying precise and correct behaviors, and then subdividing what I need, and then telling the AI to make those subdivisions works much better than trying to get the AI to bake the whole cake all at once. I still ultimately feel I am best responsible for mixing the ingredients, but the AI tells me the optimal bake time and brings me the ingredients, which is super cool I think.
Horse training being rendered obsolete by combustion engines is different than having a form of artistic expression that has been with us since we were living in caves being usurped by a machine. There’s an objective, quantifiable value to progress in a conveyance machine. But art is different.
Good take, but even then I think any former way of doing things that was more challenging and required more skill nonetheless was another thing that enriched the human experience.
A work of art is considered "good" mostly for the story it tells, it might be the creator's, or even its own story. It's usually called a "soul". Computer-generated art is souless and disgusting. Like microwaved instant pizza.
On the contrary, why should art be limited to a commericial product, isn't art beyond that of capitalism? Art will remain, it will always exists, but as a product perhaps no. Individuals will always be free to express themselves and share with others despite not making a living out of it. The idea of making a living from art is an extremely recent invention, and definitely did not exists for hundreds of thousands of years.
@@Unigma because skilled arts are somewhat hard to learn and you might as well charge for your services. The cool thing about commercial art is also that it disciplines the artist into focusing on what other people want for art and design, to expand their horizons. Also if there's enough money and the project it's big enough it's a big excuse for a team to develop which can be a rich social experience in itself. Also it's just cool as any other job to be needed and relied on in yet another field. If AI comes in and replaces most of those jobs, yet another human involved trade will die off. There'll be less opportunity for humans to interact in the ancient necessary act of commerce, an act which itself is a great excuse to dress, go out and socialize with strangers.
@@derpysean1072 I disagree - most professionally created artwork isn't made for its own sake, and as such, "good" is relative to use case. Some use cases for art don't really require a soul, and these may be automatable. (Though regardless, I feel that AI is unlikely to replace certain artforms (e.g. vector art) entirely anytime soon.)
AI won't replace coders soon. That will take about 5 years or so. So you still have some time to enjoy yourself and to find solace in videos like this one.
Actually I still believe scaling up language model leads to emergen abilities, just like what you see from GPT3. However, I think it is safe to presume OpenAI has crawl the whole Internet, so there's no more data too keep the scaling rush going(cos the scaling law is linear, extra parameters need extra data to prevent overfitting ).
They do exhibit emergent abilities as the model parameters get scaled up, in addition to the size, quality and variations of data used in pre-training. Yes OpenAI has many datasets, including the larger web crawls. They don't and won't have "All the data", this is not possible. Not to mention that much data is junk and doesn't possess useful properties. Specifically properties in line with the scaling of size and quality of the data. Scaling is not linear. Look up Chinchilla Scaling. Scaling depends on the amount of tokens, the uniqueness of the dataset and the size of the model. This is not linear.
@@Phasma6969 What do you mean scaling is not linear, it is in the abstract of the Chinchilla paper. But for OpenAI they used way more tokens than the scaling law optimal frontier. In terms of quality, that's what LLaMA and other open source models striving to have, but it just boils down to filtering, so you are not going to get more of them.
The scaling rush has a limiting factor, inference of such large models is getting out of hand fast (yes, inference! Training costs a lot but it's the inference that got really expensive really fast).
There are ways to generate more fake data though. They have already shown that generated data is still useful to train neural networks. We can also improve the way they process the data, and make them learn not just from text but with video and audio and moving in a robot body in the real world.
AI will make everyone feel like a genius making memes and hypermedia experiences like excel makes everyone feel like a genius getting around pesky human calculators when in reality, it made climate folks feel haughty about how this warming wasn't happening when it turned out they used the wrong avg function
The only thing I disagree with or would need expounded is the statement that AI can't code... we were given access to Copilot at work a few months ago, and installed the plugin in my jetbrains ide, and it was able to suggest exactly what I was going to do in a new function, class, etc. In one crazy instance, I was making a jinja template (python html templating) and was even templating the names of javascript variables inside the jinja for loop, and it gave me 100+ lines of the complete solution, minus literally one little spot where I fixed it. Could I have written that code out and typed it in the next 15 minutes with the same accuracy? Sure. But don't think that the state of AI coding tools are that they output garbage and are years away, that's just not the complete picture of reality as it stands. A year ago, I would agree... not now. Anyway. Be careful not to bury your head in the sand too much from how fast things are advancing, like I was. I had messed with self-hostable AI tooling on a semi monthly basis for the last few years... from image to text/chat to code, so I thought I had a good idea. Having said all of that - it's not going to take all the jobs. Humans will be required in many portions of the process for at least a few years, then slowly less and less, but also the humans' jobs will change for those that remain.
It depends on complexity of program you need I guess. Some framework template is a lot simpler than any specialized algorithm, e.g simulation or graph algorithm or anything else really what you can't usually do without reading science paper.
@TheIncredibleAverage either expert troll or you're an elitist that (wrongly) judges someone's skill by their language they work in, or you're afraid, or maybe all of the above. In any case you're wrong. I've had first hand experience with copilot in writing c and assembly. It has literally all of github as training fodder, but also it closely works with one's repository. So the larger the repo generally the better it is. Also used it to translate larger php and python projects into other languages. It can do a ton of heavy lifting. Do I like, endorse or trust Microsoft/github/copilot? No. Is it worthless or just a fad? Thinking that way won't help you.
@@metalstarver642 its less about complexity and more about if something has been done before. Your algorithm remark is a great point. But that still leaves 99.9% of scenarios in software out there where that's just not a problem. People have already written amazing code in so many ways, and there are countless ways that ai can help in the moment as you are writing your code as a result, at the micro level but also macro.
But dude ... They are already combining different AIs with different specialities to work together . I cant remember where I heard this but there was someone speaking about integrating an AI specialized in logic into the language AI so the ChatGPT is NOT spewing bulls**t. I think you are underestimating the AI developers a bit here . Its not like they dont understand the problem and dont know what the next step is ...
My favorite people who do art never got into it for the money. It's never been a lucrative job for money. It's always been risky. And we'll all be doing it anyway for fun.
AI art is for the most part bad and uninteresting though. The exception being the QR monster stuff where a pattern or image is embedded in another one, but the interesting part about that is the human decisions made not the generated filler in it
Llms can be trained to generate good prompts for images. As John said they're already good at bullshitting, creativity is just bullshitting in a way that other people find interesting. No humans required at all.
I disagree with his dismissive attitude of mediocre artists. Just like with programming, art, including the boilerplate commercial art, is still way a person can be paid for the timeless human act of expression through. I'd just like to know what he'd think if an ai that could also make indie ganes with complex puzzles and from-scratch dev tools. What should people do then? What kind of world are we letting be built?
Good code is a technically solvable problem by "AI" because the entire problem space is represented in a manner accessible to the "AI" I don't think jobs will be replaced any time soon though. Think about code libraries --- "oh no if there's a library for everything what good am I?" --- now there is a library for most things, and implementing them is a job requiring expertise in itself Art is a different story. Even some future "AI" tech that many deem "general intelligence" can never access artistic intuition. "AI" might be capable of excellent copies, and have access to a certain amount of re-arranging, but it will never create something new.
@@TeslasMoustache419 it's more a question of possible than easy for example, something we credit as brilliant programming like Carmack's fast inverse square root from Quake3, is a solution achievable by an LLM like system. All the moving parts and success conditions are able to be modelled, they are "visible" to AI. something like "human readable" code, or "well architected" code, not so much, because those categories are vague to AI. But finding the most efficient code is doable (at least the barrier to this is complexity). New art, on the other hand, is impossible. Look at something recent like the evolution of film. AI trained on human creation and/or cannibalizing its own creation cannot get there. There is no way for it to innovate the way a human artist can. A human artist working with AI as a tool, sure, but you need the human intuition as input or else the AI art is impossible. that's what I'm talking about more than a judgement of LLM's current ability to spit out code/fake paintings
@@chrisc7265 I agree that the models are nothing without training data made by humans in the first place, but that is true not just for art but all other domains too.
Invoking stuff like "artistic intuition" is just people desperately wanting to feel special. We're not. Our brain is a complex machine but machine nonetheless, there's no magic. It receives input and produces output, there's nothing a non-living machine couldn't also do. Same goes for creating something new.
@@digitalspecter humans clearly are special, it's an odd quirk of our time to say we aren't but besides the point, even if what you say is correct the problem of artistic intuition is not even understood, and the relative inputs are not mapped to a form readable by AI, whereas the problem of good programming is already 100% understood and already 100% readable by computers
in my thinking JB is wrong. AI will replace jobs, what before needed 10 to do, now will need 5 or 2. the point of "new tecnology takes obsolete jobs and create new jobs" is also wrong in my vision: whats the point in advancing tecnology and eficiency if the total amount of work done is about the same before and after? surely new jobs appear but the total amount of work reduces, leading to big unemployment rates. i also think JB is wrong in the matter scaling will not transform AI into super AI because it needs something diferent. AI is a language model but also is human mind, we learn by absorving whats done by our environment and reorder things in our favor, it just happens AI is still inferior in terms of processing and data than human minds
There is somewhat of a middle ground between "AI is going to obsolete all the coding jobs in a year" and "AI will never code usable software". I think, that first of all, writing complex code correctly is hard. LLMs today have no clue if their output is correct and no way to test it internally. That means if we use AI tools to code, we become specification communicators and testers of the outcome. This is something we already do in our jobs. Both of these jobs are hard and not likely to be taken over by AI any time soon. Second, there is also the high likelihood that AI will (or does already) generate the "easy" 80% of code reliably, and leaves the hard 20% to us. That is how it is with most automation tools in whatever job market you can think of. Which leads us to a world where AI, just like Copilot for you, is just another tool in the toolbox for developers to get rid of some tedious stuff. No to little layoffs needed.
When 1 person can do what 10 could have done previously, it won't lead to unemployment it will lead to 10x productivity of the whole society. On the other hand if 0 people can do what 10 could do before, it might lead to humans becoming obsolete. Our economic system assumes people's labor will always be in demand, but it will not be true anymore.
Sorry but this is an insufferably dumb take which I had to pause after about 30sec. No horse trainer had any input in the development of ICE but many, most, all if rounded up artists' work was used to train AI. Their work is weaponized against them now. We are now trying to automate not the dangerous or boring and tedious aspects of work but the very things that are BEST in us.
The AI art debate is not as simple as he makes it seem. More and more control is being given to the artist that use AI art as a tool every week. It's not fair to say some people are mediocre artists and will die because of AI art. You can't just put people in boxes and assume they'll stay in it. Some artists will be more affected by AI art than others, but the most skillful artists and those who decide to embrace AI art as a controllable tool will thrive and survive economically. I have made money with the combination of my "mediocre" art skills as you like to put it, with the help of AI as a tool. People are right to be mad about AI art. They probably had dreams of a cool career, and now it's crushed. The only thing that will remotely stop the fire is regulations, but nothing can fix the situation at this point. The best thing to do is to embrace its existance, learn how it works, and use it where it applies.
I think the debate is exactly as simple as he makes it seem. Those who fail to adapt to prevailing market conditions (i.e. use the new tools to become increasingly useful as artists) will be less likely to be profitable in the market. Not only is that not a problem, it's a good thing. It should be the case that more useful people, receive more rewards (perhaps not in principle, but definitely in practice). As an aside, I am not an artist but I thought that AI art has minimal use for most professional artists at this point. Am I wrong? I figured it can be useful for people who work on commissions for private individuals but it's not like you can generate art with anything remotely close to the consistency of style required for an actual commercial product.
@@moonboots1003well, get ready to be automated out of life at some point. It's impossible to outcompete with the machine and it is coming for 60-80% of all jobs. So it's more about your willingness or unwillingness to become human compost for the seeds of Silicon Valley/ whoever up there can afford to be an Effective Accelerationist.
@@moonboots1003also, yeah, you are wrong. Because most clients, or even just any human, who is uncreative, does not want anything "specific". Just what passes for "good enought" and is fast and takes low effort. Ai Gen gor all those boxes ticket. You could be Michelangelo, Davinci , Doré and Frazetta all in one and it'd still be like trying to stop a tsunami with a bucket. An action we could romanticize, but because new tech also alters our sensibilities to bend around its effects, I'm afraid no one would care to look. Never has there been a tool whose output so vastly surpasses the effort of input whilst so generalized in result. Expect in a few years for society never to be the same, for most people will become economically expendable sans a few culturally engrained work rituals. And wont have much where else to go. Things will get worse than get better. The whole situation of artists with AI is a microcosm of what'll happen to everyone.
@@BinaryDood I don't agree with this prediction of progress. However, I hope that you're right, I would rather not have a job and the technological advancement necessary for me to be replaced would mean that the world would be a much much much better place than it is now and there would be much more interesting things to do than work even an interesting job. Of course, you have a more pessimisstic view but I don't think we're likely to convince each other.
@@BinaryDood I don't agree here. The art for most commercial products requires consistency. You can't generate an image of Kratos from the God of War series and then in the next scene have him look pretty much the same way, he needs to actually look the same way.
GPT4 can code though, I think Blow is just thinking it can't be used for a full/real project but for simple javascript it 100% works, and it can even be used to problem solve its own errors. Obviously its far from a proper programmer but its wrong to say it can't program at all.
it can code already solved problems and slight variations of them, ever tried asking something extremely semantically specific that would be impossible to search on google already?
He is right this hysteria happened with computer drawing tools like max maya softimage blender, people made same outcry now the new drawing tool makes it so your click will be even bigger n good artists wont be affected artists can adapt went from rubbing crap on a wall to computer manipulation???
The whole concept from artists about "no one is allowed to learn from my art" is so weird to me as a programmer, not gonna lie. Nobody learns in a vacuum, we all stand on the shoulders of giants. Stop pretending y'all don't use reference art when you draw something. Nobody has given you consent to trace their art in Photoshop either, yet it's a part of every artist's workflow.
The problem isn't learning. The problem is selling it, and that happens already, en masse - AI companies will sell such art literally for pennies. And it's the same as real artists in this regard - if you train yourself on tracing, nobody cares. If you trace and sell without telling, and you're found out, you can be pretty sure lots of people will call you out for that. There have been manga artists who lost good paying jobs for that. You might say that AI just creates new stuff after learning - but that's not true either. AI takes pieces of their learning material and combines them. It doesn't (and currently cannot) start from a blank canvas, the finished product is inherently a collage (based off of some algorithm), not a new work.
a person using someone else's art as reference for practice is not the same as a company scraping every image on the internet to feed into a machine that will eventually disrupt the global economy.
Yes but also no. It's like porn, the real thing is much better but a lot of people won't even try given the proxy. AI is already at a college student level of specialized knowledge in most fields. It went from zero to that in 2-3 years and that's with a "cold start". AI begets more powerful AI, and as with most things in tech, it's an exponential curve. Also literally 100% of people who start doing complicated things like coding, suck at it for a long time. If it takes years of experience to outperform an AI, then most people won't have the kind of discipline to do something years at a time without any payoff.
The [important] part that people miss in this topic is that artistic jobs (not only painting) are one of the few where people are into them because they love them first and foremost, and only second as a means of making money. There is some of that in programming too, rarely in other things. But most jobs people would not do for free. They would art. Just think of that popular idea from years back, that if robots did everything for us, we would sit back and do art. And AI kills that. Yes you still can, but it's now a problem. Don't want to waste my time explaining why in a yt comment.
If you want to do art for free, AI can not challenge that in any way. Machines are made to replace jobs, not hobbies. People right now still enjoy all the hobbies that as jobs have long been delegated to factories - brewing soaps, sewing plush toys, cooking, pottery and probably hundreds of other things, too.
@@containercore6832 Your argument makes zero sense. I've seen much more competent art from people who love to make art instead of those who use art as just means to an end. And that's no anecdote of my own - there is this little thing called "Overjustification Effect" you might want to look into.
@@sophiasabkv1372 It makes perfect sense that a person who spends their whole workday committed to the study of the art achieves more than someone who can only do it during their after-work leisure hours (at the expense of other activities such as socializing or exercise). The overjustification effect is completely unrelated to this. Nowhere did I say that commerce is the driving motivation behind good art. Let me reiterate "The lyf so short, the craft so long to lerne".
@@containercore6832 How is it "completely unrelated" when it directly shuts down the silly idea that hobbies produce art of amateur quality unless they double as jobs? Throwing around some cool quotes in Latin isn't gonna make your argument sound.
artists fearing AI art over what big companies and governments are trying to do with it scares me a bit more... It's being sold as the best thing ever to solve all world problems, while some of their outputs sound worse than a toddler's. In reality, much of AI right now is a toy, and normal people are being asked to play with it to get free testing. It's also being gladly used as a target for kneejerk draconian laws and regulations. Because AI is just big computing, I wouldn't be surprised if someone says we need to start having computer control...
It incentives people to be removed from the process of creation behind things. It's an implied level of displacement so high it might rival the industrial revolution. Honestly a radical social shift will have to happen to stalemate narrow and holistic values in regards to this tech, so we dont throw ourselves into a world where the output of these things blur out reality and puts people out of jobs infinitely faster than "new jobs" get to be created or can pivot into other occupations. I personally think creativity should only be automated in a post-scarcity society, if ever. Creativity is in thinking itself, it's bigger than just art. We dont need more "content", if anything we need less for saturation is so paralyzing already without AI. I think this is a far more existencial topic than people are allowing themselves to admit.
Also it definitely will be used as an excuse for absolute "human authentication" by big data rofl. Every move you do to be saved up so it can distinguish you from endless artificial noise. A human barcode of sorts. Made by the same institutions who created the problem in the first place. This is pretty predictable really. C2PA and Sam Altman's UBI cryptocurrency eye scanning thing clearly show their motivations.
@@BinaryDood part of the problem is also people thinking "well the industrial revolution was good, so this will be fine too!" But the industrial revolution caused a lot of the lifestyle problems we face today. Terrible work hours, jobs that feel meaningless and depressing, feeling disconnected from your community, generalists being made obsolete, etc. are all caused by the industrial revolution because skilled trades and local businesses were destroyed by automation. We haven't even solved these issues as a society yet and we're potentially on the precipice of a new revolution? Doesn't sound promising.
@@NihongoWakannai I wouldn't even say the industrial revolution is what caused all of those issues alone. The tech brought on by the internet has exacerbated the issue for certain, but I'd say having voice conversations is a social activity. Lurking on a forum isn't really. Social aspects are much more complex there, and people trying to mix them with AI are just popping another algorithm to solve a very complex problem that has no one solution.
@@NihongoWakannai What is lost on a lot of the people rn who willingly disregard the displaced is that back in the IR, people with their livelyhoods at stake didnt magically just travel 20 years in time to enjoy the positive outcomes of the industrial revolution without its immediate downsides... No, they had to fight tooth and nail to force the big owners to pay attention to their life-or-death plight. A lot of societal upheaval was necessary to get to even out a baseline which woul later cultivate the means to a welfare state. They didnt lie on the floor waiting to die because "it's just progress. It's liable for people to become obsolete! What?! Just look at our shovels! You saying you'd rather be a hand digger?" Plus, we are far more fragile now because of intercontinental economic dependencies becoming so much more intricate. Plus, gen AI is highly generalist and of lower bar of entry than any previous technology (normally they are either one or the other, and to far less extents on each). Sooner or later, things are going to hit the hard line of material reality and it's gonna become apartment that we(they) opened Pandora's Box.
I don't think this is a good take, especially because many concerns around AI and art not even being mentioned here: - People's work being used without consent. - People not being compensated for the value their work gives to these generative AIs. - The blurred distinguishability between AI works and human works, which is only going to get more prominent over time. - That we're still in early stages of this technology, and it will become better at catering to specific prompts in the future (won't just be a randomish output as suggested in the video). - The general way these tools lead to the cheapening of artistic expression. For someone who talks about finding meaning in hard work, I think it's sad to see such dismissal of artist's concerns. If we're talking about the functionality and quality of output of today's generative AIs are, then yes it's not so concerning, but they are obviously going to improve from here.
Artwork can already be used to create new artwork without consent or compensation when a human artist uses it as a reference, inspiration or a learning tool. Why do you think it's an issue if it becomes difficult to distinguish between 'human works' and 'AI works'? "That we're still in early stages of this technology, and it will become better at catering to specific prompts in the future (won't just be a randomish output as suggested in the video)." Why do you believe this? Why would it be a problem? Of course it's possible for the technology to develop in unexpected ways however these models work stochastically and therefore I wouldn't necessarily expect that they will develop the ability to produce art with enough consistency of style to be broadly useful in commercial contexts, at least to the extent of replacing artists. Behind all this, I just don't see why it's a problem. The lack of consent and compensation stuff is obviously an issue if taken seriously, but I don't see how it's different from the way in which human artists use other peoples art without consent or compensation. More generally though, isn't it a good thing when new tools become available to make artists more productive? Even if generative AI does develop to be broadly useful for professional artists working on products, won't we just produce more art and higher quality art than we did before? Isn't that an awesome thing? So, commercial art will be produced differently. Commercial art is already produced differently than it was 20 years ago. And 20 years before that it was produced differently still.
@@moonboots1003 When it comes to making art, the generative AIs of today do not learn to make it in any way like a human. Distinguishing between human and AI works is important to those of us who actually want to see meaningful things in our lives (and this goes beyond art as well). Generally, I want to see what a human thinks and feels, I don't care to see some mashup generated by a machine, however convincing it may be. If I see someone's drawing, or someone writes me a email / letter, I'm only interested if it's personal. Outside of maybe joking around, a generated work may as well be meaningless. Now when an Artificial General Intelligence comes along, I may be interested in what it produces - and it may even produce some real art! - but I want to know that the AGI made it, that it was the AGI's expression, not a human taking credit for it. This is my personal view, but I think more and more people will feel like this as this continues. Right now, there's a general sentiment Google is becoming more useless. Machine generated articles that optimize SEO get to the top of search results, and these writings are hollow, soulless and often pointless and a waste of time. This is why, to me, it's important now and only going to become more important. The world is already inauthentic enough. AI is already being used to replace artists. These tools can already do things decently well, especially things in specific styles. They can do a good job illustrating backgrounds, and anime characters, psychedelic animations etc. And this is only the early stages. This is why the consent and compensation is so important. Artists deserve to be compensated, because without them this generative AI wouldn't be able to exist. AI companies and investors are very aware of this by the way, they know the value of their products depends highly on the works they have used. Yes, the way art is produced will continue to change, that change does not mean people don't deserve compensation for their works. I don't agree that AI will make commercial art (nor other kinds) higher quality in general, but that's subjective - and in some selective cases, it might do, such as videogames. For many other areas though, I just think it will just cheapen art. I'll just add this. There are currently more books, music, films, videogames than you could possibly experience in one lifetime. I'd argue you can't even experience all the offerings that are well worth experiencing in a single lifetime, and that's aside from all the other things you can experience in life as well! What value does adding all this machine generated work add to the world? If a human does a high quality illustration today, it might take them 20 hours. They'll put something of their soul in that work. When I see an illustration today, even if the quality isn't that good, I can feel there was purpose behind it. What purpose will be behind the instantly generated illustrations that are going to (already starting to) flood the world, and how will we find what matters amongst the sea of meaninglessness?
@@MarkPTP7000 I see your point with the first paragraph, and while it's a shame if this is taken away from you, it's your personal preference and not something that requires regulation. But don't you find it extremely unlikely that AI work will become so indistinguishable from non-AI work that you won't be able to find and appreciate non-AI work if you wish? As long as there are people like you who put special value into the work that isn't created using AI, there will be a market for that work. It seems that you generally have a low opinion of the real usefulness of generative AI in its current state. This is a reasonable opinion and so I wonder why you think it's a big problem. It isn't currently very useful, so how is it an issue? The hollow soulless crap can be ignored and will die as the hype dies. I don't really see how a human using someone's art to help them learn how to make art without consent or compensation is different from a human using someone's art to help them learn how to train a model. I could be convinced otherwise but then companies would just end up paying licence fees for banks of legal-to-use training data, AI would still develop in the same way. And that's assuming that the use of training data from the internet is enforceable which I really don't think it is (and it's pointless to make regulations that aren't enforceable). So I'm not sure it makes much of a difference. Artists don't stand to make any notable money from this sort of deal anyway, just look at Spotify, and AI has nothing to do with that. I find it hard to believe that AI is as useful as you say at replacing artists... Sure it can show you 'An anime character' but it cannot create an anime character and then recreate the same character consistently with the exact same style which is what is necessary for the vast majority of commercial products. Okay true, it can probably make some pretty patterns for greeting cards, but if your only ability as an artist was to make pretty patterns for greeting cards, now you aren't useful anymore so why should you expect to have a job without gaining new skills? Even if you're compensated for your role in training the AI model that's going to make you almost literally nothing. You have to be actually useful for someone to pay you. With your last paragraph, I see where you're coming from but it seems obvious to me that as the pool of total works increases, the quality of the highest quality works also tends to increase. If AI makes any sort of worker more productive, that worker will then be able to spend more time on different aspects of developing the product and therefore make a better product. If a concept artist can use generative AI to mock up some inspirational references more quickly, then she can spend more time considering the best style for the game and more time on creating the final reference pieces. Most works are not created whole cloth from the hearts and souls of auteur geniuses, they are business processes and like any other commercial product they stand to increase in quality by any and all improvements in workflow efficiency. If it becomes possible for a worker to use AI to consistently and precisely produce art to the level that is required for a commercial product which has a consistent creative style then either 1) It's basically just a new way of drawing. If the human artist can use AI to produce EXACTLY what he has in his head, then he is still the artist and the expression is his own. He doesn't have to have good hand-eye coordination, but if you think the soul of art is about the fact that the artist had good hand-eye coordination I don't know what to say. 2) Or the AI will be capable of this because it is actually intelligent and then it is a person so none of your concerns are relevant. Biological persons are not superior to digital persons. Otherwise, it's going to remain only able to generate vague stuff because of its stochastic core and is not gonna be much of a problem for the majority of skilled artists doing anything useful as part of a team.
@@moonboots1003 Answering your questions here not to seem rude or blunt, just to answer quicker than a massive essay again. I did not mention regulation about distinguishing between human and AI works. That said, it's actually one of the main areas legislators are already looking at, with talk of watermarking systems. I don't think that will work in the long run. 'But don't you find it extremely unlikely that AI work will become so indistinguishable from non-AI work that you won't be able to find and appreciate non-AI work if you wish?' For certain art styles this is already the case! And I only think this tech will get better in this regard. ' As long as there are people like you who put special value into the work that isn't created using AI, there will be a market for that work' It is already hard to distinguish for some kinds of art. The illustration community on twitter already deal with this problem quite often. 'It seems that you generally have a low opinion of the real usefulness of generative AI in its current state. ' In terms of producing creations that are interesting / worthwhile yes I do. As commercial products, they are already useful. My opinion of them as hollow does not mean they are useless. AI art is already being used in commercial products. 'I don't really see how a human using someone's art to help them learn how to make art without consent or compensation is different from a human using someone's art to help them learn how to train a model. I could be convinced otherwise...' It is very different, I would have to write for an age to explain properly. I suggest you look into how they work, and see if you feel the same after. 'Artists don't stand to make any notable money from this sort of deal anyway, just look at Spotify, and AI has nothing to do with that.' That's defeatist, and does not need to be true. I suspect it probably will be though because AI companies will have more power and influence to make things go their way. What happens in the early stage will set the precedent. And even if it is only a little money, so be it. Right now many exploitative business models of tech are being pushed against (huge percentage fees in the app stores). Just because tech has the advantage does not mean that has to be the case. Music and film companies were what forced the internet to use copyright detection. They had the power to do so. Smaller artists and businesses don't have that, but they may have a chance for something similar. 'Sure it can show you 'An anime character' but it cannot create an anime character and then recreate the same character consistently with the exact same style which is what is necessary for the vast majority of commercial products.' It can do that right now. It's not simple, but it's not that hard either, and that's why so called 'AI Prompters' are saying prompting these models is a skill. It will get much easier and better at producing consistent results soon. I don't disagree that AI can't be used to make better products, nor do I think it's necessarily bad if used as a tool in the process (e.g Hey AI, colour this character's overalls blue in every frame in this movie). But it's not 'basically just a new way of drawing' .The AI does not digitally make brush strokes. It's image generation. 'Or the AI will be capable of this because it is actually intelligent and then it is a person so none of your concerns are relevant. Biological persons are not superior to digital persons.' I agree. That is what most refer to as Artificial General intelligence, which I (and pretty much most experts) do not believe the current AIs are. AGI will in my opinion, be a real new form of life. I will be more interested in what art THAT makes, even though a lot of problems mentioned will still be in play. As I mentioned, I don't want to see humans passing work that was made by AIs as their own. I don't want the world spammed with low effort, meaningless garbage. We will probably desire human only spaces when AGI happens though, just because AGI will be able to produce vastly more and in tiny amounts of time. By the way, I know a lot of what happens with AI is going to be inevitable. I'm actually not all anti-AI. The potential greatness it could bring to the world is enormous. I just think in terms of creativity, and exploitation of creative people, it's got some very sad effects, and we don't actually have to just let AI companies get away with whatever they want.
Finally someone says it, Jonathan is so full of contradictions on so many things, languishing the loss of human expression then telling artists to learn to code, you really should only listen to him when it comes to programming and even then with a grain of salt, it's also tiring to see people pretend like artists are gatekeeping when they just want to protect they're hard work, we're talking about individuals here not multimillion dollar companies with a patent.
@@shroomcraftgamesIt's already at my craft (programming) and I'll use it and ride the wave and evolve my occupation as such. Being a neo-luddite is silly.
@@xevious4142 Their work was not required to build AI. Their work was required to build AI that does art. AI that does art is less important than AI that does things like maximize the strength/weight ratio of a 3d printed part. People have emotions and art does have an impact on quality of life. But its a lot less of an impact than things that help people not die. A little art goes a long ways. Its an oversaturated market. Art can be duplicated and shared across the world. We don't need 50% of the population to be doing some form of art.
Again, he's having very strong opinions about something he's far from an expert on. To be fair there is wide disagreement about how or if LLMs will continue to scale, but we have very accurate scaling laws for the loss of these models, and when the loss goes down we tend to have emergent abilities. Those scaling laws have definitely not come to an end yet, so we're likely to be surprised yet again by the next model and next. He's right this doesn't lead to intelligence in the same way as humans approach problems, but then again multimodal LLMs can be trained on thousands of lifetimes worth of text, images and video, and we will eventually be able to put whole codebases in context, so I think it's a bit short sighted to claim these types of models will never be able to code well. I was initially a skeptic, but you can only be surprised so many times before you have to update your own world model.
Even the language is changing. In fields where deprecation of old methods is common (CS), or even critical (medicine, safety, ...), it's bad to retain old data in the model. While emergent abilities are nice, it would be much better if the decision making part was delegated to something that gives 100% correct (or no) choices instead of "filling in the blanks" with the closest match. Assuming a company feeds and ideal base model with all its IP and best practices, the model will still only be able to generate a mixture of those or something "creative" that likely doesn't work. If you feed the "creative" thing into a test suite you end up with a brute force approach that provides you with infinite garbage you need to shift through in order to get to something valuable. They do be good at writing boilerplate though.
@@thewiseowl8804 I've worked in ML and AI for 13 years, I think that makes me more qualified than him on this particular topic. He's literally done no AI work in his life, except some 90s classes at Berkley perhaps, so that bar is not particularly high to beat. I admit he's smarter than me, so if he spent some focused time working on actual AI, then sure...
The argument about old jobs that are no longer relevant doesn't hold any water when the starting prompt is that these advancements are built on the backs of the prpduction of artists that did not consent to their work being used to fuel their replacements, or used at all. That is something that did not have to happen with simple economy of efficiencies that made those jobs go away, all of this production is only possible on the basis of the availability of training data. And the "learn to code" meme is so pointless, especially when software development is probably going to be the field most affected by these developments, with millions of jobs either disappearing overnight or being reduced to massaging shitty AI prompts and output rather than doing any actual software development. Whether those jobs are valuable or not is a whole different topic, but people need to eat.
the idea that anyone should be able to sit on their copyright to this extent, like a dragon on its treasure, is massively detrimental to society I understand very well the difficulty of monetizing creativity, but this is not the answer.
The whole idea that this is based on artists work is wrong from the start. The image generators are trained on billions of completely random junk images from the Web, not art specifically. They learn how to make images from that and what image feature go together with words. Once that initial training is done you can teach them a new style or character with a handful of examples and a few minutes of GPU time on your PC at home. If you'd remove all the artists images from the training set, nothing would change, everybody could still steal your art style simply by using an image as input instead of a word. Not much different than taking an image from one artist and telling another one "paint me something like that".
@@chrisc7265 the extent of which has been not at all? all these ai companies just took every single piece of art every single person has ever made, gave them absolutely nothing, and now is using it to make all of them lose their jobs. the only thing you won't get the ai to generate for you is movie transcripts because movie companies have 7000 laywers that cost a billion a year. no average person is allowed to "hoard" or even own anything.
I doubt any author of any text whose work you've read in your lifetime would consent to having their text used as training data, only to result in such an uninformed comment.
AI is NOT Art is a tool that generates images from real artists' works, photos and pictures by merging. A Human Artist is always in process of learning for this reason there aren't mediocre artists as you said for ignorance. For instance there are professional artists and amateur artists that range from masterful to unpolished to beginner. Do you understand?
What about the fax machine repairmen? Surely we should ban the internet, so that they fax machine repair guys can keep their jobs. Honestly Stable Diffusion (and similar tech) are just another tool in an artist's box, if they are good at what they do. It's no different than stuff like google image search (for finding a picture of anything) or the "undo button" with drawing/tablet software, etc. Just tools that can elevate the work of good artists.
@@BinaryDood What do you mean? I don't understand your reply. I can't even tell if you're disagreeing with me? The printing press made text more accessible, the uber made getting home from the bar more accessible. The wheel made horses more accessible. I'm not sure what you're getting at. The AI models (like stable diff) produce garbage. It's useful garbage but it's still garbage. If there are low-skilled artists who can't compete with ai-gen garbage then they're in the wrong industry.
Yeah, except an undo button or image search can't fully replace whatever value you can produce as an artist. Eventually image gen will be good enough that you will not need to supervise it, so the artist will become completely obsolete.
@@SergeeeekI don't think so, because AI can't make hand-made art or paint a literal canvas. The AI has unlimited technical skill, but is limited in terms of creativity and innovation. Artists used to fear that the printing press would be the end of their craft, or the photocopier/printer.
@@torgo_ you're right, we're left only with physical mediums for the time being. But arguably the volume of digital art vs physical art produced today leans on the side of digital. It's easier to sell, easier to share etc.
what if low artists and mediocer artist and good artists made use of this. so they are now using what normal non artists use. with the knowledge they have accumalted and the hours they have spent trying to get good at doing art, they will match up with good artists. but the thing is that good artists will be even good. if not good then faster. so it's a win win. I don't understand why ppl are so mad about this
It's a loose-loose. You are using the system which can exist without your usage of it. Floor is so low any ignorant's work indistinguishable from a pro. There are changes than only pros can do, and in a lot of cases it is already being used for that. But everyone doing so knows it's an unavoidable free admission to be ridden off altogether.
AI art isn't human expression and brings the whole work into question. The more a work is done by AI, the less of it is the artist's expression, the more of it is whatever the AI model found on DeviantArt.
Would the electrical power suppliers be able to keep up if AI became as competent as your average developer? A 10-15K dollar GPU has to consume so much power just to put out a subpar 200-character long output. Not to mention it's low capacity for context.
@@Sergeeeek because the expectation is going up. consumption is only going up. anyway, that's a huge waste of resource just to replace humans. hubris of privileged people, what can I say.
@@3_smh_3 humans aren't using 0 energy either. We just use food instead of electricity directly. The thing is that each of those AI model deployments, while inefficient, is still serving thousands of requests simultaneously. A single person can never compete with that, even if AI sucks now.
@@Sergeeeek I'd rather not have electrical outages and carbon emissions than generate images and texts. maybe not a problem where you live. Also, AI's energy efficiency will likely never reach anyhere near human's brains. Again, hubris.
@@3_smh_3 the space is blowing up, it's a matter of time before we get hyper-specialized hardware that can only do AI stuff while being much more efficient than GPUs today. Right now most of ML stuff is still implemented in software.
What is this "How do you think human artists learn?" nonsense. He contradicts himself right off the bat. First he compares AI art using hard data as "learning like a human artist would". Then says there is NO general intelligence or knowledge behind AI at all. What a delusional thing to say lol.
I see his point that the very best of the best humans will still continue to create for a long time.
However, I see a mass homogenization of all popular media in the future. Similar to how 99% of Unity/Unreal games look the same, if you provide people with an out-of-the-box easy solution, they won't bother to come up with their own unique solution. This is something that Blow talks about a lot with using frameworks.
Also, every master has to go through a journeyman phase. AI replacing the journeyman could potentially prevent a lot of human potential in the future. I don't have to worry about that thankfully but worry about future generations.
The problem is that right now you'd have to review the AI generated code to make sure it's not complete nonsense, which means you did nothing other than changing your work from writing code to reviewing code, which is more tedious. Not to mention in many cases knowing what the code is supposed to do is the actual work, so those AI systems right now are just a better intellisense in the best case scenario. Which is fine but it's not what many hyped people expect.
Completely correct. There is also the problem of application and things being adjusted for "software as a service." I have objections to SAAS but it's beside the point. I think general articial intelligence is required in order to actually run such a product because it would need a mind for the problem space.
Artwork is actually REALLY flexible when it comes to satisfying a need or application, software is a bit different.
The problem is not wether the models scale forever, but of their fast adoption rate in the socioeconomic sphere. If generative LLMs become a baseline dependency anywhere close to the internet in most markets, then I'm afraid we'll generate new sensibilities which would be "okay" with subpar output. And, because it is so much cheaper and faster, we might devalue that combo of effort+creativity to the point of it becoming not worth for anyone to pursue, as it would always hit a brick wall of endless saturation and expectations too quick for any direct human input. It becomes a race to the bottom... The middle class is fucked.
I don't know, I feel like similar change have happened in different industries before.
Like the industrial revolution, and producing as scale for cheaper. Like IKEA can be seens as producing an "okay" subpar output compared to artisanal woodcraft. All in all it doesn't seem so bad. cheap product have some place in the market.
IMO a lot of place where illustration was just not affordable will now be able to exist.
@@ZoobTheElement huge societal upheaval needed to happen during the Industrial Revolution so that people weren't just left to die. You live in the here and the now, not in 20 years in the future where you can enjoy the positive outcomes of this tech's implementation without suffering its immediate direct results on a socioeconomic sphere not ready for it. You too have a stake in this. Also there shall be so much of artificial "content" that, unlike something made in the material world, shall completely blur out any modicum of reality out of the picture. Also, art, or creativity in general, does not have a direct assigned functionality in order for a bare minimum to be met and mass produced forever, it is manifest in the process itself and molded by the wills of those making it. Creativity is one of the most important aspects of our thinking and it should not be outsource en mass. You can't have a whole society of individuals who dont know how to do anything yet are entitled to feel themselves the creators of endless products which they know nothing off of the process behind each of their supposed fields.
Uncles Ted's view of technology seems to be getting more and more validation
@@SkeleTonHammer nope, not necessarily
This makes the assumption that the average product produced by humans is better than what can be produced by a generative LLMs.
I can't speak to other domains, but when it comes to coding common patterns, ChatGPT 4.0 often produces results that have better structure than the average developer -- assuming you understand the context to use the outputs. This is because these models have weighting and selected training to choose sources that are better than what you might find if you sampled a random developer.
Are there examples of bad output? Sure, and for anecdotes that you can find about bad AI you can find analogous examples in people; something that people tend to discredit when discussing generative models.
Now, as Jon said, above average talent will be in demand for a long time. Having the ability to think, utilize context, and perform better the norm will be valuable as long as AGI doesn't exist.
Is the middle class fucked? Probably within the near future, and that's largely a function of AI being able to replicate common patterns and work flows, which constitute the majority of middle class work.
Should we as a society figure out how to support our community as we find ways to replace their jobs? Absolutely, and there will likely be a lot of growing pains with this shift in society.
Very good take from JB once again. I don't consider myself a good programmer, but I can't imagine myself using AI to help with anything beyond small and/or tedious tasks. If I were to order it to make a whole program, whatever it may be, I wouldn't be able to trust to not fuck something(s) up somewhere, and debugging that would not be a pleasant experience. This is by no means a diss at modern LLMs, as the fact that they are capable of doing even that just from a prompt in the first place is truly a miracles of science .
i think AI should not be confused with a person in the respect that it is expected to do all by itself but i also think AI can perfectly make a whole program and only be tested/judged by a real person before it enters real time production. AI should be the mean and not the end by itself. humans are the end
@@gabrielchuede6688 "and only be tested/judged by a real person" runs heavily into "and debugging that would not be a pleasant experience". Any experienced programmer will tell you that understanding others' code (and possibly refactoring it) can be times harder than writing your own code.
(And let's not even go into stuff like intentional flaws in the code which can be missed by anyone not "very" proficient in their field - I've seen stuff like encryption function that looks okay and works okay... at first, once you encrypt larger thing, it fails to encrypt, and you might not even notice. And the change compared to the normal one is tiny, one line that looks like it should work, but it doesn't, and you need great understanding of bitwise operations and encryption process to gather why it fails to begin with.
Now let's imagine your LLM goes through the internet in search of such encryption function for your program and implements this one instead of the main one. Why? Nobody knows, it's LLM!)
And the small tedious tasks are usually a decent break from those that require harder thinking.
Or so ingrained that you can do it quickly with smart use of tools and just raw muscle memory, making it effectively nothing but a neat novelty, but nothing impressive.
@@NukeCloudstalker I'd rather take an actual break.
@@samuelbucher5189 well, I'm happy that my overarching point got across well enough, that the only contention was whether "getting less hard work done between hard work" or "taking a total break" is better.
I’m so happy you’re uploading again!
Also a human artist doesn’t necessarily train himself on other artists’ work. They can see something in the real world and then in their minds imagine a way to paint that, all without needing to train on someone else’s work
And, you know, 5 senses. Decoded by real time abstract thinking. Everyone runs into this basic heuristic contradiction. It's just so easy.
sure, they can imagine and paint that.... AFTER having trained. Just like AI.
Yep, the best training for the visual arts has always been still life and figure drawing.
My wife is a digital artist. She still has more commission work than she can handle. AI just isn't on her level and can't do the stuff she does (AI doesn't understand layers, which are necessary for some work such as high-level vtuber models that have 100+ animated layers). It does, however, help with references. Think of how Pinterest works. That is what can be automated nicely for artists. Tons of references which are good images, but not consistent with what you asked (and full of small problems that would prevent them from being usable as-is).
Isn't it essential for a reference to be accurate?
If I wanted to draw one, why would I look at an AI tiger where the perspective is just completely messed up and the anatomy weirdly shifted?
@@hexcodeff6624 that's the thing, not at all. For example, some newer AIs have better prompt following and consistency than the SD 1.5 based one she likes, even if the image is 512x768 she doesn't care, she likes those ones. No matter if the hands are messed up, or if the background doesn't make too much sense; she doesn't trace the image or anything - as long as some details look cool it gets her creative juices flowing. She has no problem drawing perfect hands by herself, so AI is more of a "throw me some ideas".
@@4.0.4 That's not a reference then, at least not in the way I understand it.
@@hexcodeff6624 her "actual" references also aren't like the final product. Maybe she liked the pattern on a plaid or the way light was hitting a character. She uses 3D stuff like DesignDoll or CSP's for anatomy.
@@4.0.4 Yeah, I just don't think reference is the right word for the AI stuff here.
Inspiration, maybe. It is, after all, not much more than a glorified Image search algorithm.
As a artist I kind of agree about the notion of work, but at the same time its hard to not notice the stark contrast between. Oh AI wont be able to program in a long time while also saying, I did not expect to see AI art to be so good in my own lifetime.
AI has an "outrunning the bear" problem. It doesn't need to be a general intelligence, it just has to convince enough humans that it's a general intelligence.
This shits fucking terrifying
It doesn't have to do that. It just has to produce something of value.
But if it’s not a general intelligence it’s unlikely it would be able to do that in the first place
oh, like humans applying for a job
im glad he can say it's more than he expected in his lifetime. i agree! i love to live in this surprising time
I only disagree with his stance on how far we are with coding. Yeah, it's nowhere near perfect, but there are still so many mediocre programmers who make everything they work on worse, at which point having it done by an AI that explicitly broadcasts what it's doing is just... more sensible. Not that it matters, because all that changes in a matter of months anyway. But still, most professions have shortcomings we could just abstract away.
Generative art depends mostly on computer vision than general intelligence.
Jon will change his mind on this once the AI starts making puzzle games
@TheIncredibleAverageusing an AI to assist with art should bring into question whether or not the whole work was made by AI entirely.
@henriquemarques6196 If there are no jobs then you won't need a job... How is this a problem? We do jobs in order to produce the things that we as humans need and want, so if there are no jobs, we will have the things we need and want produced by AI.
@@moonboots1003 Only in theory. Some major societal shifts need to happen to accommodate such a change, though (e.g. UBI).
AI art has an abject banal mad magazine aesthetic to it. That's the problem they have a latent quality common to everything they output and it's subtly off-putting. If people saturate media with those images it's not good. Using chat gpt to quickly look up documentation though is useful.
I google "Victorian Architecture" and most of the first image results where Ai generated. I could tell, dunno if the average person could. Unchecked, this could become really devious. A lot of articles I've had the feeling that they ar e AI generated, but I can't tell.
@@BinaryDoodhopefully this continues and the internet is drawn to its conclusion as a medium too open to robots for its own good.
Bring back books, catalogues, human curation etc.
@@krunkle5136 yes!
@@krunkle5136 The lack of curation and editorship is one of the main issues with the internet, it can't be done by a robot. Glad that other people see it too.
@@krunkle5136 It's people using AI to circumvent IP ownership, not robots.
People will sell books and catalogues and what not generated by AI for the same reasons they post it online.
we need hundreds of thousands of jobs for bad artists for the field to be strong enough to profuce a few actually good outliers. Good artists are a product of a system, if the skill ceiling gets raised so high you will only get hired if you're one of the top specialiats in your field, it's doomed.
Not hypothetical, I've used AI to code. It helped me get started in a project. I ended up rewriting everything it did and taking a different approach. I would have been probably faster without it but I was stuck, so it did help.
People are way too scared. It's just a tool. A bit eerie, but the sky isn't falling.
similar experience, using chatgpt4 definitely helps me start and unblock me in some projects, even though I end up rewriting most if not everything, it's also great for getting some feedback on what some obfuscated code is doing if I'm not seeing it immediately
it don't work on lowlevel stuffs actually its terrible if you try to do something that has not been done before or is not that present on the web they start hallucinating . To express my taught i think for now the language models are wrong they are not designed to work with hard problems and especially novel ones, they have to come up with other models that are not manly driven by knowledge more about generating new patterns from existing ones even though they do that in some way it's still too slow and inefficient still requiring a generalization of the pattern you aim for . I would also like to add to my previous statements that even in high level tasks they still hallucinate if you go out of the norms and trends thus you have to take it with a grain of salt . That being said they still have their uses as you just pointed it out it's great to get you started or search a general information i think they do it better than a search engine in certain cases .
The problem with AI isn't the jobs, the problem is copyright theft, blatant infringement on rights.
Depends on what you mean by theft. If you publish something without saying anything, it can legitimally be used to train a real artist and a neural network. If you explicitly forbid it from being fed to machines (and most art wasn't), then if you agree with intellectual property rights then yeah by all means just ban that stuff, but that won't stop the changes.
Jobs are infinite. As a matter of fact, "jobs" is a term used by the status quo to denote what the financial and political sector of a society wants to be done, having a general idea of who can perform what based on past experience. If a politician or banker decides that making holes in his garden is a worthy endeveour for his amusement, he will hire a company expert in hole diging and they themselves will hire low skilled workers to shovel soil out of the garden if needed (for customized hole shapes a generic driller can't perform).
Any idea why Jon has stopped putting out stream vods? Was it to do with people uploading clips?
I think it's one of those things where it's just up to artists to gatekeep AI from their space it if they wanna gatekeep. The main thing I see potentially becoming more of a thing with AI art is more of low skilled artists using AI to generate images and then just manipulate them in photoshop to sorta cheat their way to somewhat decent looking art quick that they otherwise wouldn't really be able to draw at their skill level. I think stuff like that should be faced with a ton of gatekeeping from the art community, and I think it generally has been.
I always enjoy Jon's thinking
But why care about computer stuff while we let human life languish- bad politics, bad culture, pointless consumerism etc
The problem isn't artists losing their jobs, the problem is why we still need jobs to survive in an increasingly automated society. We should be excited about AI replacing human labor; only in capitalism is the loss of jobs a death sentence for everyone who gets replaced. We can't keep "making new jobs" forever once AGI and advanced robotics are on earth. Humans will simply be unemployable in every industry. Under capitalism, that means 99% of the population starving to death.
Think what you will about capitalism, but it is quickly becoming outdated.
This. AI taking our jobs should be a good thing. It should mean Great! I don't need to work any more and still retain the same lifestyle!
@@steve16384but you won't. Who will pay you to just live without participating in the economy in any way?
@@BinaryDood Something like Global Basic Income. If the AI and automation is generating wealth for very little cost, the price of items should be a lot cheaper, meaning GBI needn't be prohibitively expensive for society. See countries with good social welfare already.
People need to involve themselves in activities that benefit their community and require skill and effort or else people atrophy.
People need work, ideally work where they're not mistreated (much of post industrial service jobs are mistreatment of the soul).
Anyone saying otherwise should check with someone to make sure they don't have major depression.
@@krunkle5136 that's very humane of you! I agree pretty much with everything you said. I do like tech, but sometimes most of its possible permeations are highly disrupting and chaotic.
People are going to have to realize that, even outside of AI, many of the job markets will be unreliable in the future. Technology changes the jobs available, and that has always been true. The problem is that we've tied people's ability to consume too tightly to their ability to produce something unique, so when a machine can produce what a person can, suddenly the person has lost the ability to consume. You COULD control how people make art, but then you're creating a convoluted system to artificially keep employment up. You're also limiting the creative output of what a writer can make with a computer, which might be violating the creative rights of the writer.
Perhaps the solution to these employment problems will eventually be the government paying for continuing education for adults that suddenly need to switch jobs. Of course people may also need a UBI if they have a family and need to go back to school. The whole thing is very complicated. But again, it's probably foolish to try to hold back technology for the purpose of keeping jobs. You are more likely to pass legislation for a stronger social safety net than you are to pass legislation limiting what a computer can observe.
And yes, people need more empathy for anyone feeling that their jobs could be replaced by a machine, but "is it good that this happens" is a separate discussion from "will this happen".
What's that version of solitaire he is playing?
did you find out ? edit: I did (at the start or screenshot + image search)
Zachtronics Solitare Collection.
@@captainnoyaux No, not yet.
The name of the game is at the start of the video.
I keep playing this video in a loop so I can solve the solitaire puzzle in the background... 😅😅😅
What is the game he's playing?
Teach a man how to fish:
Step 1. Screenshot it
Step 2. Upload it to Google image search.
Step 3. Eat the fish.
Literally comes up at the start of the video.
He’s playing "Zachtronics Solitare Collection." I’m sorry you had to deal with these two awful, unhelpful robots.
It is true that the name of the game is at the start of the video.
Well, here's the thing: he's a professional programmer, and Jonathan correctly states that AI cannot solve programing. He does this because he is knowledgeable about the topic.
But Jon is not an artist, yet he confidently states that we shouldn't listen to Artists, then proceeds with a weird philosopher king analogy involving horse trainers. He then callously states that he doesn't care if Artists (or whatever he refers to as "low skill" and "mediocre") don't have jobs. And as before, in a topic he knows nothing about, he again proceeds with another weird analogy, this time about coal miners.
It is fine to want to have an authoritative opinion on most things, but the problem is that he painfully doesn't know about what he is talking about in some topics, yet he proceeds with an uninformed opinion anyways. Even though all things considered, his first answer applies pretty well to both Art and programming.
Artists arent horses. Art isnt about leading you from point A to point B. If you continue with that analogy then you are also willing for a lot of people to die, as that was the case for horses. It's weird. One of the main themes of The Witness to me was precisely that. The obsession of the process itself, translating the surrounding world with abstract thought to then draw back the line onto it. It really makes me sad. Seems like its not the way he thought the "collapse of civilization" would happen, though it absolutely fits the broader vision he implied (because yes, not just artists are on this tech's line of sight).
@rodrigoptunio spot on. That you for posting this
This really isn't an uninformed take. He believes:
- People becoming redundant due to new technology is inevitable and fine, you just need to give them enough time to reskill
- People already learn art by accessing free resources en masse and mimicking it until you can't recognise their exact sources, generative AI does this at scale with the only difference being that it's a bot not a human
You can disagree, but these are reasonable and informed stances.
@@MrLordFireDragon there are places where redundancy should never be applied. Like outsourcing all creativity. It is in thinking itself, not just "art as job". Anyone who can see more than a palm in front of them and isnt convenientely narrowviewing the issue, would know the chaotic ramifications of this. Also, huge societal upheaval is often downright needed to even out such level of displacement, so countless (literally experts here) arent left to die by being cast aside whilst the very thing they made is used aganist them infinitely, with no say on the mather.
And that is untrue. Read Harold Speed to realize that is an insuficient euristic. Humans must first breakdown their own utilitary symbolic interpretations of the world which the mind decodes from other senses that arent vision. Unlearning the method of signs to learn the method of seeing. The organic process of seeing the natural world from reference, or other art, is not an input -(blackbox)- output. Much less of countless finished readymade tobe consumed outputs. It requires your senses, your conscious learning, physicality of the process, recreation as opposed to merging (white canvas as opposed to noise) and a stake in life altogether, which guides your actions. You are thinking that the map comes before the place.
@@MrLordFireDragon the idea that what ai is doing is akin to what a human being is doing when viewing art is complete horseshit. You probably have next to no experience with making art just like Jonathan.
He didn’t say he doesn’t care if they have jobs. He outright answers that question by saying they will do other jobs.
Being an artist should allow you to get a job and at the very least you shouldn't be disposed for an algorithm copying the stuff you made.
Not caring that Jr. Artists will be replaced sort of goes against his whole "collapse of civilization" narrative. Art is a stronghold of culture and therefore civilizations. To blindly give it up to automation seems fatalistic.
Love the guy but as an artist, this one caught me in the fee fees 😢
yeah Jon has some odd, bitter takes these days..
1. If junior artists being replaced is furthering the collapse of civilization, then it would be supporting his idea of collapse of civilization
2. the copyright/capitalist model has been massively destructive to society (not only artists), and things like this are just exposing a flaw that already existed. I think this stuff needs to burn itself to the ground before something better can grow from the ashes, this is just how the cycle of civilizations goes.
3. artists actually have a ton of opportunity now, relatively speaking. Indeed to continue with the "collapse of civilization" theme, the collapse is when opportunity really opens up. AI spitting out these tacky images (that will become tackier as more people get used to AI art and it is classified as "fake" and "lame") is not a real threat to anyone with ambition. Those with artistic sense will see AI as another tool in their kit.
@@chrisc7265 I agree with most of what you said. What I've been saying is the "hand-drawn" artist will be more valuable in the coming years than AI models. Similar to a luthier who makes guitars by hand; you can go to Guitar Center and pick up a manufactured product for $200 or pay someone 9K to do it by hand. The human element will always be valuable.
I've noticed that when interacting with "AI stuff" Jon Blow apparently ceases to believe his "collapse of civilization" talk and most of his takes are conflicting and were addressed back then. (the talk even have an AI related question, I've checked).
I find this phenomena very weird.
Industry "artists" are not artists anyway, they are industry workers, which is why their jobs will be automated. They are a cog in a machine waiting to be changed to a fancier and better cog. What I would call actual artists serve a completely different purpose and they will be fine either with or without AI.
Just thought of a good one: Johnathan Blowfish
Most people aren't that smart so the bar has been set very low for the AI agents and surpassing it shouldn't make the AI agents break a sweat.
Current AI models have no way of getting experience like humans do, so their "intellegence", if it is ever spawned, would be quirky and not reason nor flake at reasoning like humans.
There's a lot of misconceptions in this video. For example, language models are counterintuitive, because they don't actually use language.
What they are is text prediction algorithms. For question answering, they just process text from websites like reddit. Because the they're trained on question answering, they are able to answer questions, by predicting the answer.
You can use more complex methods or "agentic workflows" to break down and process complicated problems in the same way. Unlike basic chatbots, these make use of reasoning loops (thinking out loud, in text form) and simplified outputs (like JSON objects/methods)
You seem to be the one with misconceptions.
In order to get good at predicting text, they DID have to learn some language stuff. It's naive to say they don't use language. They use it (even if not perfectly) to predict text.
They don't process text for question answering. They don't have "real" text stored anythwere, they just have a neural network. It has text in the same way that you have images in your memory.
@@neopabo No you can't. LLMs (or neural networks in general) are capable of producing useful answers in a way that was not possible before. No better way to do these things has been shown, which doesn't use neural networks.
@@MrTomyCJI was thinking about how to answer you but honestly it's too complicated.
So, believe what you will.
What about the street horse poop cleaners!
The code being output for huge programs is vapid nonsense, but code for individual modules is on point
I find that being able to spend more time architechting and specifying precise and correct behaviors, and then subdividing what I need, and then telling the AI to make those subdivisions works much better than trying to get the AI to bake the whole cake all at once.
I still ultimately feel I am best responsible for mixing the ingredients, but the AI tells me the optimal bake time and brings me the ingredients, which is super cool I think.
Horse training being rendered obsolete by combustion engines is different than having a form of artistic expression that has been with us since we were living in caves being usurped by a machine. There’s an objective, quantifiable value to progress in a conveyance machine. But art is different.
Good take, but even then I think any former way of doing things that was more challenging and required more skill nonetheless was another thing that enriched the human experience.
A work of art is considered "good" mostly for the story it tells, it might be the creator's, or even its own story. It's usually called a "soul".
Computer-generated art is souless and disgusting. Like microwaved instant pizza.
On the contrary, why should art be limited to a commericial product, isn't art beyond that of capitalism? Art will remain, it will always exists, but as a product perhaps no. Individuals will always be free to express themselves and share with others despite not making a living out of it.
The idea of making a living from art is an extremely recent invention, and definitely did not exists for hundreds of thousands of years.
@@Unigma because skilled arts are somewhat hard to learn and you might as well charge for your services.
The cool thing about commercial art is also that it disciplines the artist into focusing on what other people want for art and design, to expand their horizons.
Also if there's enough money and the project it's big enough it's a big excuse for a team to develop which can be a rich social experience in itself.
Also it's just cool as any other job to be needed and relied on in yet another field.
If AI comes in and replaces most of those jobs, yet another human involved trade will die off. There'll be less opportunity for humans to interact in the ancient necessary act of commerce, an act which itself is a great excuse to dress, go out and socialize with strangers.
@@derpysean1072 I disagree - most professionally created artwork isn't made for its own sake, and as such, "good" is relative to use case. Some use cases for art don't really require a soul, and these may be automatable. (Though regardless, I feel that AI is unlikely to replace certain artforms (e.g. vector art) entirely anytime soon.)
AI won't replace coders soon. That will take about 5 years or so. So you still have some time to enjoy yourself and to find solace in videos like this one.
it will never
Actually I still believe scaling up language model leads to emergen abilities, just like what you see from GPT3. However, I think it is safe to presume OpenAI has crawl the whole Internet, so there's no more data too keep the scaling rush going(cos the scaling law is linear, extra parameters need extra data to prevent overfitting ).
They do exhibit emergent abilities as the model parameters get scaled up, in addition to the size, quality and variations of data used in pre-training. Yes OpenAI has many datasets, including the larger web crawls. They don't and won't have "All the data", this is not possible. Not to mention that much data is junk and doesn't possess useful properties. Specifically properties in line with the scaling of size and quality of the data. Scaling is not linear. Look up Chinchilla Scaling. Scaling depends on the amount of tokens, the uniqueness of the dataset and the size of the model. This is not linear.
@@Phasma6969 What do you mean scaling is not linear, it is in the abstract of the Chinchilla paper. But for OpenAI they used way more tokens than the scaling law optimal frontier. In terms of quality, that's what LLaMA and other open source models striving to have, but it just boils down to filtering, so you are not going to get more of them.
The scaling rush has a limiting factor, inference of such large models is getting out of hand fast (yes, inference! Training costs a lot but it's the inference that got really expensive really fast).
There are ways to generate more fake data though. They have already shown that generated data is still useful to train neural networks. We can also improve the way they process the data, and make them learn not just from text but with video and audio and moving in a robot body in the real world.
AI will make everyone feel like a genius making memes and hypermedia experiences like excel makes everyone feel like a genius getting around pesky human calculators when in reality, it made climate folks feel haughty about how this warming wasn't happening when it turned out they used the wrong avg function
The only thing I disagree with or would need expounded is the statement that AI can't code... we were given access to Copilot at work a few months ago, and installed the plugin in my jetbrains ide, and it was able to suggest exactly what I was going to do in a new function, class, etc. In one crazy instance, I was making a jinja template (python html templating) and was even templating the names of javascript variables inside the jinja for loop, and it gave me 100+ lines of the complete solution, minus literally one little spot where I fixed it.
Could I have written that code out and typed it in the next 15 minutes with the same accuracy? Sure. But don't think that the state of AI coding tools are that they output garbage and are years away, that's just not the complete picture of reality as it stands. A year ago, I would agree... not now.
Anyway. Be careful not to bury your head in the sand too much from how fast things are advancing, like I was. I had messed with self-hostable AI tooling on a semi monthly basis for the last few years... from image to text/chat to code, so I thought I had a good idea.
Having said all of that - it's not going to take all the jobs. Humans will be required in many portions of the process for at least a few years, then slowly less and less, but also the humans' jobs will change for those that remain.
It depends on complexity of program you need I guess. Some framework template is a lot simpler than any specialized algorithm, e.g simulation or graph algorithm or anything else really what you can't usually do without reading science paper.
@TheIncredibleAverage either expert troll or you're an elitist that (wrongly) judges someone's skill by their language they work in, or you're afraid, or maybe all of the above. In any case you're wrong.
I've had first hand experience with copilot in writing c and assembly. It has literally all of github as training fodder, but also it closely works with one's repository. So the larger the repo generally the better it is.
Also used it to translate larger php and python projects into other languages. It can do a ton of heavy lifting.
Do I like, endorse or trust Microsoft/github/copilot? No. Is it worthless or just a fad? Thinking that way won't help you.
@@metalstarver642 its less about complexity and more about if something has been done before. Your algorithm remark is a great point.
But that still leaves 99.9% of scenarios in software out there where that's just not a problem. People have already written amazing code in so many ways, and there are countless ways that ai can help in the moment as you are writing your code as a result, at the micro level but also macro.
Programmers are not safe either
Jon spitting rational facts, as always. I love him so much.
But dude ...
They are already combining different AIs with different specialities to work together .
I cant remember where I heard this but there was someone speaking about integrating an AI specialized in logic into the language AI so the ChatGPT is NOT spewing bulls**t.
I think you are underestimating the AI developers a bit here .
Its not like they dont understand the problem and dont know what the next step is ...
My favorite people who do art never got into it for the money. It's never been a lucrative job for money. It's always been risky. And we'll all be doing it anyway for fun.
AI art is for the most part bad and uninteresting though. The exception being the QR monster stuff where a pattern or image is embedded in another one, but the interesting part about that is the human decisions made not the generated filler in it
it's because it doesn't have an intention, but the user expresses the intention as a prompt.
Generic prompts get you generic images.
Llms can be trained to generate good prompts for images. As John said they're already good at bullshitting, creativity is just bullshitting in a way that other people find interesting. No humans required at all.
What game is this?
hes right
I disagree with his dismissive attitude of mediocre artists. Just like with programming, art, including the boilerplate commercial art, is still way a person can be paid for the timeless human act of expression through.
I'd just like to know what he'd think if an ai that could also make indie ganes with complex puzzles and from-scratch dev tools. What should people do then? What kind of world are we letting be built?
Good code is a technically solvable problem by "AI" because the entire problem space is represented in a manner accessible to the "AI"
I don't think jobs will be replaced any time soon though. Think about code libraries --- "oh no if there's a library for everything what good am I?" --- now there is a library for most things, and implementing them is a job requiring expertise in itself
Art is a different story. Even some future "AI" tech that many deem "general intelligence" can never access artistic intuition. "AI" might be capable of excellent copies, and have access to a certain amount of re-arranging, but it will never create something new.
I have the opposite view, I think making art (painting) is much easier for AI than software engineering/programming.
@@TeslasMoustache419 it's more a question of possible than easy
for example, something we credit as brilliant programming like Carmack's fast inverse square root from Quake3, is a solution achievable by an LLM like system. All the moving parts and success conditions are able to be modelled, they are "visible" to AI.
something like "human readable" code, or "well architected" code, not so much, because those categories are vague to AI. But finding the most efficient code is doable (at least the barrier to this is complexity).
New art, on the other hand, is impossible. Look at something recent like the evolution of film. AI trained on human creation and/or cannibalizing its own creation cannot get there. There is no way for it to innovate the way a human artist can. A human artist working with AI as a tool, sure, but you need the human intuition as input or else the AI art is impossible.
that's what I'm talking about more than a judgement of LLM's current ability to spit out code/fake paintings
@@chrisc7265 I agree that the models are nothing without training data made by humans in the first place, but that is true not just for art but all other domains too.
Invoking stuff like "artistic intuition" is just people desperately wanting to feel special. We're not. Our brain is a complex machine but machine nonetheless, there's no magic. It receives input and produces output, there's nothing a non-living machine couldn't also do. Same goes for creating something new.
@@digitalspecter humans clearly are special, it's an odd quirk of our time to say we aren't
but besides the point, even if what you say is correct the problem of artistic intuition is not even understood, and the relative inputs are not mapped to a form readable by AI, whereas the problem of good programming is already 100% understood and already 100% readable by computers
"Tell them to learn to code" an absolutely nuclear take lol
Artists will be fine. Content creators... Maybe not.
A lot of artists are content creators maybe most.
you know when he had to say it was amazing he died a little inside
in my thinking JB is wrong. AI will replace jobs, what before needed 10 to do, now will need 5 or 2. the point of "new tecnology takes obsolete jobs and create new jobs" is also wrong in my vision: whats the point in advancing tecnology and eficiency if the total amount of work done is about the same before and after? surely new jobs appear but the total amount of work reduces, leading to big unemployment rates.
i also think JB is wrong in the matter scaling will not transform AI into super AI because it needs something diferent. AI is a language model but also is human mind, we learn by absorving whats done by our environment and reorder things in our favor, it just happens AI is still inferior in terms of processing and data than human minds
There is somewhat of a middle ground between "AI is going to obsolete all the coding jobs in a year" and "AI will never code usable software".
I think, that first of all, writing complex code correctly is hard. LLMs today have no clue if their output is correct and no way to test it internally. That means if we use AI tools to code, we become specification communicators and testers of the outcome. This is something we already do in our jobs. Both of these jobs are hard and not likely to be taken over by AI any time soon.
Second, there is also the high likelihood that AI will (or does already) generate the "easy" 80% of code reliably, and leaves the hard 20% to us. That is how it is with most automation tools in whatever job market you can think of. Which leads us to a world where AI, just like Copilot for you, is just another tool in the toolbox for developers to get rid of some tedious stuff. No to little layoffs needed.
When 1 person can do what 10 could have done previously, it won't lead to unemployment it will lead to 10x productivity of the whole society.
On the other hand if 0 people can do what 10 could do before, it might lead to humans becoming obsolete. Our economic system assumes people's labor will always be in demand, but it will not be true anymore.
Sorry but this is an insufferably dumb take which I had to pause after about 30sec. No horse trainer had any input in the development of ICE but many, most, all if rounded up artists' work was used to train AI. Their work is weaponized against them now. We are now trying to automate not the dangerous or boring and tedious aspects of work but the very things that are BEST in us.
The AI art debate is not as simple as he makes it seem.
More and more control is being given to the artist that use AI art as a tool every week.
It's not fair to say some people are mediocre artists and will die because of AI art. You can't just put people in boxes and assume they'll stay in it.
Some artists will be more affected by AI art than others, but the most skillful artists and those who decide to embrace AI art as a controllable tool will thrive and survive economically.
I have made money with the combination of my "mediocre" art skills as you like to put it, with the help of AI as a tool.
People are right to be mad about AI art. They probably had dreams of a cool career, and now it's crushed. The only thing that will remotely stop the fire is regulations, but nothing can fix the situation at this point. The best thing to do is to embrace its existance, learn how it works, and use it where it applies.
I think the debate is exactly as simple as he makes it seem.
Those who fail to adapt to prevailing market conditions (i.e. use the new tools to become increasingly useful as artists) will be less likely to be profitable in the market. Not only is that not a problem, it's a good thing. It should be the case that more useful people, receive more rewards (perhaps not in principle, but definitely in practice).
As an aside, I am not an artist but I thought that AI art has minimal use for most professional artists at this point. Am I wrong? I figured it can be useful for people who work on commissions for private individuals but it's not like you can generate art with anything remotely close to the consistency of style required for an actual commercial product.
@@moonboots1003well, get ready to be automated out of life at some point. It's impossible to outcompete with the machine and it is coming for 60-80% of all jobs. So it's more about your willingness or unwillingness to become human compost for the seeds of Silicon Valley/ whoever up there can afford to be an Effective Accelerationist.
@@moonboots1003also, yeah, you are wrong. Because most clients, or even just any human, who is uncreative, does not want anything "specific". Just what passes for "good enought" and is fast and takes low effort. Ai Gen gor all those boxes ticket. You could be Michelangelo, Davinci , Doré and Frazetta all in one and it'd still be like trying to stop a tsunami with a bucket. An action we could romanticize, but because new tech also alters our sensibilities to bend around its effects, I'm afraid no one would care to look. Never has there been a tool whose output so vastly surpasses the effort of input whilst so generalized in result. Expect in a few years for society never to be the same, for most people will become economically expendable sans a few culturally engrained work rituals. And wont have much where else to go. Things will get worse than get better. The whole situation of artists with AI is a microcosm of what'll happen to everyone.
@@BinaryDood I don't agree with this prediction of progress. However, I hope that you're right, I would rather not have a job and the technological advancement necessary for me to be replaced would mean that the world would be a much much much better place than it is now and there would be much more interesting things to do than work even an interesting job.
Of course, you have a more pessimisstic view but I don't think we're likely to convince each other.
@@BinaryDood I don't agree here. The art for most commercial products requires consistency. You can't generate an image of Kratos from the God of War series and then in the next scene have him look pretty much the same way, he needs to actually look the same way.
GPT4 can code though, I think Blow is just thinking it can't be used for a full/real project but for simple javascript it 100% works, and it can even be used to problem solve its own errors. Obviously its far from a proper programmer but its wrong to say it can't program at all.
it can code already solved problems and slight variations of them, ever tried asking something extremely semantically specific that would be impossible to search on google already?
GPT4 fails at fresh hard leetcode problems for a reason
He is right this hysteria happened with computer drawing tools like max maya softimage blender, people made same outcry now the new drawing tool makes it so your click will be even bigger n good artists wont be affected artists can adapt went from rubbing crap on a wall to computer manipulation???
The whole concept from artists about "no one is allowed to learn from my art" is so weird to me as a programmer, not gonna lie. Nobody learns in a vacuum, we all stand on the shoulders of giants. Stop pretending y'all don't use reference art when you draw something. Nobody has given you consent to trace their art in Photoshop either, yet it's a part of every artist's workflow.
You constructed a beautiful strawman. You should be an artist.
The problem isn't learning. The problem is selling it, and that happens already, en masse - AI companies will sell such art literally for pennies.
And it's the same as real artists in this regard - if you train yourself on tracing, nobody cares. If you trace and sell without telling, and you're found out, you can be pretty sure lots of people will call you out for that. There have been manga artists who lost good paying jobs for that.
You might say that AI just creates new stuff after learning - but that's not true either. AI takes pieces of their learning material and combines them. It doesn't (and currently cannot) start from a blank canvas, the finished product is inherently a collage (based off of some algorithm), not a new work.
a person using someone else's art as reference for practice is not the same as a company scraping every image on the internet to feed into a machine that will eventually disrupt the global economy.
No one is telling you, a person, to not learn from their art. Machine learning isn't the same as human learning
Only the worst artists are protective of their art style.
Yes but also no. It's like porn, the real thing is much better but a lot of people won't even try given the proxy.
AI is already at a college student level of specialized knowledge in most fields. It went from zero to that in 2-3 years and that's with a "cold start". AI begets more powerful AI, and as with most things in tech, it's an exponential curve.
Also literally 100% of people who start doing complicated things like coding, suck at it for a long time. If it takes years of experience to outperform an AI, then most people won't have the kind of discipline to do something years at a time without any payoff.
The [important] part that people miss in this topic is that artistic jobs (not only painting) are one of the few where people are into them because they love them first and foremost, and only second as a means of making money. There is some of that in programming too, rarely in other things. But most jobs people would not do for free. They would art. Just think of that popular idea from years back, that if robots did everything for us, we would sit back and do art. And AI kills that. Yes you still can, but it's now a problem. Don't want to waste my time explaining why in a yt comment.
If you want to do art for free, AI can not challenge that in any way. Machines are made to replace jobs, not hobbies. People right now still enjoy all the hobbies that as jobs have long been delegated to factories - brewing soaps, sewing plush toys, cooking, pottery and probably hundreds of other things, too.
@@sophiasabkv1372 Then it's relegated to being only a hobby for amateurs. Remember, "Ars longa, vita brevis".
@@containercore6832 Your argument makes zero sense. I've seen much more competent art from people who love to make art instead of those who use art as just means to an end. And that's no anecdote of my own - there is this little thing called "Overjustification Effect" you might want to look into.
@@sophiasabkv1372 It makes perfect sense that a person who spends their whole workday committed to the study of the art achieves more than someone who can only do it during their after-work leisure hours (at the expense of other activities such as socializing or exercise). The overjustification effect is completely unrelated to this. Nowhere did I say that commerce is the driving motivation behind good art. Let me reiterate "The lyf so short, the craft so long to lerne".
@@containercore6832 How is it "completely unrelated" when it directly shuts down the silly idea that hobbies produce art of amateur quality unless they double as jobs? Throwing around some cool quotes in Latin isn't gonna make your argument sound.
artists fearing AI art over what big companies and governments are trying to do with it scares me a bit more... It's being sold as the best thing ever to solve all world problems, while some of their outputs sound worse than a toddler's.
In reality, much of AI right now is a toy, and normal people are being asked to play with it to get free testing. It's also being gladly used as a target for kneejerk draconian laws and regulations. Because AI is just big computing, I wouldn't be surprised if someone says we need to start having computer control...
It incentives people to be removed from the process of creation behind things. It's an implied level of displacement so high it might rival the industrial revolution. Honestly a radical social shift will have to happen to stalemate narrow and holistic values in regards to this tech, so we dont throw ourselves into a world where the output of these things blur out reality and puts people out of jobs infinitely faster than "new jobs" get to be created or can pivot into other occupations. I personally think creativity should only be automated in a post-scarcity society, if ever. Creativity is in thinking itself, it's bigger than just art. We dont need more "content", if anything we need less for saturation is so paralyzing already without AI. I think this is a far more existencial topic than people are allowing themselves to admit.
Also it definitely will be used as an excuse for absolute "human authentication" by big data rofl. Every move you do to be saved up so it can distinguish you from endless artificial noise. A human barcode of sorts. Made by the same institutions who created the problem in the first place. This is pretty predictable really. C2PA and Sam Altman's UBI cryptocurrency eye scanning thing clearly show their motivations.
@@BinaryDood part of the problem is also people thinking "well the industrial revolution was good, so this will be fine too!" But the industrial revolution caused a lot of the lifestyle problems we face today. Terrible work hours, jobs that feel meaningless and depressing, feeling disconnected from your community, generalists being made obsolete, etc. are all caused by the industrial revolution because skilled trades and local businesses were destroyed by automation.
We haven't even solved these issues as a society yet and we're potentially on the precipice of a new revolution? Doesn't sound promising.
@@NihongoWakannai I wouldn't even say the industrial revolution is what caused all of those issues alone. The tech brought on by the internet has exacerbated the issue for certain, but I'd say having voice conversations is a social activity.
Lurking on a forum isn't really.
Social aspects are much more complex there, and people trying to mix them with AI are just popping another algorithm to solve a very complex problem that has no one solution.
@@NihongoWakannai What is lost on a lot of the people rn who willingly disregard the displaced is that back in the IR, people with their livelyhoods at stake didnt magically just travel 20 years in time to enjoy the positive outcomes of the industrial revolution without its immediate downsides... No, they had to fight tooth and nail to force the big owners to pay attention to their life-or-death plight. A lot of societal upheaval was necessary to get to even out a baseline which woul later cultivate the means to a welfare state. They didnt lie on the floor waiting to die because "it's just progress. It's liable for people to become obsolete! What?! Just look at our shovels! You saying you'd rather be a hand digger?" Plus, we are far more fragile now because of intercontinental economic dependencies becoming so much more intricate. Plus, gen AI is highly generalist and of lower bar of entry than any previous technology (normally they are either one or the other, and to far less extents on each). Sooner or later, things are going to hit the hard line of material reality and it's gonna become apartment that we(they) opened Pandora's Box.
I don't think this is a good take, especially because many concerns around AI and art not even being mentioned here:
- People's work being used without consent.
- People not being compensated for the value their work gives to these generative AIs.
- The blurred distinguishability between AI works and human works, which is only going to get more prominent over time.
- That we're still in early stages of this technology, and it will become better at catering to specific prompts in the future (won't just be a randomish output as suggested in the video).
- The general way these tools lead to the cheapening of artistic expression.
For someone who talks about finding meaning in hard work, I think it's sad to see such dismissal of artist's concerns. If we're talking about the functionality and quality of output of today's generative AIs are, then yes it's not so concerning, but they are obviously going to improve from here.
Artwork can already be used to create new artwork without consent or compensation when a human artist uses it as a reference, inspiration or a learning tool.
Why do you think it's an issue if it becomes difficult to distinguish between 'human works' and 'AI works'?
"That we're still in early stages of this technology, and it will become better at catering to specific prompts in the future (won't just be a randomish output as suggested in the video)." Why do you believe this? Why would it be a problem? Of course it's possible for the technology to develop in unexpected ways however these models work stochastically and therefore I wouldn't necessarily expect that they will develop the ability to produce art with enough consistency of style to be broadly useful in commercial contexts, at least to the extent of replacing artists.
Behind all this, I just don't see why it's a problem. The lack of consent and compensation stuff is obviously an issue if taken seriously, but I don't see how it's different from the way in which human artists use other peoples art without consent or compensation. More generally though, isn't it a good thing when new tools become available to make artists more productive? Even if generative AI does develop to be broadly useful for professional artists working on products, won't we just produce more art and higher quality art than we did before? Isn't that an awesome thing?
So, commercial art will be produced differently. Commercial art is already produced differently than it was 20 years ago. And 20 years before that it was produced differently still.
@@moonboots1003 When it comes to making art, the generative AIs of today do not learn to make it in any way like a human.
Distinguishing between human and AI works is important to those of us who actually want to see meaningful things in our lives (and this goes beyond art as well). Generally, I want to see what a human thinks and feels, I don't care to see some mashup generated by a machine, however convincing it may be. If I see someone's drawing, or someone writes me a email / letter, I'm only interested if it's personal. Outside of maybe joking around, a generated work may as well be meaningless. Now when an Artificial General Intelligence comes along, I may be interested in what it produces - and it may even produce some real art! - but I want to know that the AGI made it, that it was the AGI's expression, not a human taking credit for it. This is my personal view, but I think more and more people will feel like this as this continues. Right now, there's a general sentiment Google is becoming more useless. Machine generated articles that optimize SEO get to the top of search results, and these writings are hollow, soulless and often pointless and a waste of time. This is why, to me, it's important now and only going to become more important. The world is already inauthentic enough.
AI is already being used to replace artists. These tools can already do things decently well, especially things in specific styles. They can do a good job illustrating backgrounds, and anime characters, psychedelic animations etc. And this is only the early stages. This is why the consent and compensation is so important. Artists deserve to be compensated, because without them this generative AI wouldn't be able to exist. AI companies and investors are very aware of this by the way, they know the value of their products depends highly on the works they have used.
Yes, the way art is produced will continue to change, that change does not mean people don't deserve compensation for their works. I don't agree that AI will make commercial art (nor other kinds) higher quality in general, but that's subjective - and in some selective cases, it might do, such as videogames. For many other areas though, I just think it will just cheapen art.
I'll just add this. There are currently more books, music, films, videogames than you could possibly experience in one lifetime. I'd argue you can't even experience all the offerings that are well worth experiencing in a single lifetime, and that's aside from all the other things you can experience in life as well! What value does adding all this machine generated work add to the world? If a human does a high quality illustration today, it might take them 20 hours. They'll put something of their soul in that work. When I see an illustration today, even if the quality isn't that good, I can feel there was purpose behind it. What purpose will be behind the instantly generated illustrations that are going to (already starting to) flood the world, and how will we find what matters amongst the sea of meaninglessness?
@@MarkPTP7000
I see your point with the first paragraph, and while it's a shame if this is taken away from you, it's your personal preference and not something that requires regulation. But don't you find it extremely unlikely that AI work will become so indistinguishable from non-AI work that you won't be able to find and appreciate non-AI work if you wish? As long as there are people like you who put special value into the work that isn't created using AI, there will be a market for that work.
It seems that you generally have a low opinion of the real usefulness of generative AI in its current state. This is a reasonable opinion and so I wonder why you think it's a big problem. It isn't currently very useful, so how is it an issue? The hollow soulless crap can be ignored and will die as the hype dies.
I don't really see how a human using someone's art to help them learn how to make art without consent or compensation is different from a human using someone's art to help them learn how to train a model. I could be convinced otherwise but then companies would just end up paying licence fees for banks of legal-to-use training data, AI would still develop in the same way. And that's assuming that the use of training data from the internet is enforceable which I really don't think it is (and it's pointless to make regulations that aren't enforceable). So I'm not sure it makes much of a difference. Artists don't stand to make any notable money from this sort of deal anyway, just look at Spotify, and AI has nothing to do with that.
I find it hard to believe that AI is as useful as you say at replacing artists... Sure it can show you 'An anime character' but it cannot create an anime character and then recreate the same character consistently with the exact same style which is what is necessary for the vast majority of commercial products. Okay true, it can probably make some pretty patterns for greeting cards, but if your only ability as an artist was to make pretty patterns for greeting cards, now you aren't useful anymore so why should you expect to have a job without gaining new skills? Even if you're compensated for your role in training the AI model that's going to make you almost literally nothing. You have to be actually useful for someone to pay you.
With your last paragraph, I see where you're coming from but it seems obvious to me that as the pool of total works increases, the quality of the highest quality works also tends to increase. If AI makes any sort of worker more productive, that worker will then be able to spend more time on different aspects of developing the product and therefore make a better product. If a concept artist can use generative AI to mock up some inspirational references more quickly, then she can spend more time considering the best style for the game and more time on creating the final reference pieces. Most works are not created whole cloth from the hearts and souls of auteur geniuses, they are business processes and like any other commercial product they stand to increase in quality by any and all improvements in workflow efficiency.
If it becomes possible for a worker to use AI to consistently and precisely produce art to the level that is required for a commercial product which has a consistent creative style then either
1) It's basically just a new way of drawing. If the human artist can use AI to produce EXACTLY what he has in his head, then he is still the artist and the expression is his own. He doesn't have to have good hand-eye coordination, but if you think the soul of art is about the fact that the artist had good hand-eye coordination I don't know what to say.
2) Or the AI will be capable of this because it is actually intelligent and then it is a person so none of your concerns are relevant. Biological persons are not superior to digital persons.
Otherwise, it's going to remain only able to generate vague stuff because of its stochastic core and is not gonna be much of a problem for the majority of skilled artists doing anything useful as part of a team.
@@moonboots1003 Answering your questions here not to seem rude or blunt, just to answer quicker than a massive essay again.
I did not mention regulation about distinguishing between human and AI works. That said, it's actually one of the main areas legislators are already looking at, with talk of watermarking systems. I don't think that will work in the long run.
'But don't you find it extremely unlikely that AI work will become so indistinguishable from non-AI work that you won't be able to find and appreciate non-AI work if you wish?' For certain art styles this is already the case! And I only think this tech will get better in this regard.
' As long as there are people like you who put special value into the work that isn't created using AI, there will be a market for that work' It is already hard to distinguish for some kinds of art. The illustration community on twitter already deal with this problem quite often.
'It seems that you generally have a low opinion of the real usefulness of generative AI in its current state. ' In terms of producing creations that are interesting / worthwhile yes I do. As commercial products, they are already useful. My opinion of them as hollow does not mean they are useless. AI art is already being used in commercial products.
'I don't really see how a human using someone's art to help them learn how to make art without consent or compensation is different from a human using someone's art to help them learn how to train a model. I could be convinced otherwise...' It is very different, I would have to write for an age to explain properly. I suggest you look into how they work, and see if you feel the same after.
'Artists don't stand to make any notable money from this sort of deal anyway, just look at Spotify, and AI has nothing to do with that.' That's defeatist, and does not need to be true. I suspect it probably will be though because AI companies will have more power and influence to make things go their way. What happens in the early stage will set the precedent. And even if it is only a little money, so be it. Right now many exploitative business models of tech are being pushed against (huge percentage fees in the app stores). Just because tech has the advantage does not mean that has to be the case. Music and film companies were what forced the internet to use copyright detection. They had the power to do so. Smaller artists and businesses don't have that, but they may have a chance for something similar.
'Sure it can show you 'An anime character' but it cannot create an anime character and then recreate the same character consistently with the exact same style which is what is necessary for the vast majority of commercial products.' It can do that right now. It's not simple, but it's not that hard either, and that's why so called 'AI Prompters' are saying prompting these models is a skill. It will get much easier and better at producing consistent results soon.
I don't disagree that AI can't be used to make better products, nor do I think it's necessarily bad if used as a tool in the process (e.g Hey AI, colour this character's overalls blue in every frame in this movie). But it's not 'basically just a new way of drawing' .The AI does not digitally make brush strokes. It's image generation.
'Or the AI will be capable of this because it is actually intelligent and then it is a person so none of your concerns are relevant. Biological persons are not superior to digital persons.' I agree. That is what most refer to as Artificial General intelligence, which I (and pretty much most experts) do not believe the current AIs are. AGI will in my opinion, be a real new form of life. I will be more interested in what art THAT makes, even though a lot of problems mentioned will still be in play. As I mentioned, I don't want to see humans passing work that was made by AIs as their own. I don't want the world spammed with low effort, meaningless garbage. We will probably desire human only spaces when AGI happens though, just because AGI will be able to produce vastly more and in tiny amounts of time.
By the way, I know a lot of what happens with AI is going to be inevitable. I'm actually not all anti-AI. The potential greatness it could bring to the world is enormous. I just think in terms of creativity, and exploitation of creative people, it's got some very sad effects, and we don't actually have to just let AI companies get away with whatever they want.
Finally someone says it, Jonathan is so full of contradictions on so many things, languishing the loss of human expression then telling artists to learn to code, you really should only listen to him when it comes to programming and even then with a grain of salt, it's also tiring to see people pretend like artists are gatekeeping when they just want to protect they're hard work, we're talking about individuals here not multimillion dollar companies with a patent.
Need more good takes like these, a%tists getting too unhinged with their entitlement.
Calling artists entitled when it was their work that was required to build these systems is fuckin rich lol
Once it comes for your craft, we will see how entitled you will get
@@xevious4142they are entitled. This idea that some industries should have protection from technological advances at the cost of society is entitled.
@@shroomcraftgamesIt's already at my craft (programming) and I'll use it and ride the wave and evolve my occupation as such. Being a neo-luddite is silly.
@@xevious4142 Their work was not required to build AI. Their work was required to build AI that does art. AI that does art is less important than AI that does things like maximize the strength/weight ratio of a 3d printed part. People have emotions and art does have an impact on quality of life. But its a lot less of an impact than things that help people not die. A little art goes a long ways. Its an oversaturated market. Art can be duplicated and shared across the world. We don't need 50% of the population to be doing some form of art.
bro just went with the most boomer take and called it a day
Again, he's having very strong opinions about something he's far from an expert on. To be fair there is wide disagreement about how or if LLMs will continue to scale, but we have very accurate scaling laws for the loss of these models, and when the loss goes down we tend to have emergent abilities. Those scaling laws have definitely not come to an end yet, so we're likely to be surprised yet again by the next model and next. He's right this doesn't lead to intelligence in the same way as humans approach problems, but then again multimodal LLMs can be trained on thousands of lifetimes worth of text, images and video, and we will eventually be able to put whole codebases in context, so I think it's a bit short sighted to claim these types of models will never be able to code well. I was initially a skeptic, but you can only be surprised so many times before you have to update your own world model.
Even the language is changing. In fields where deprecation of old methods is common (CS), or even critical (medicine, safety, ...), it's bad to retain old data in the model. While emergent abilities are nice, it would be much better if the decision making part was delegated to something that gives 100% correct (or no) choices instead of "filling in the blanks" with the closest match. Assuming a company feeds and ideal base model with all its IP and best practices, the model will still only be able to generate a mixture of those or something "creative" that likely doesn't work. If you feed the "creative" thing into a test suite you end up with a brute force approach that provides you with infinite garbage you need to shift through in order to get to something valuable.
They do be good at writing boilerplate though.
He should be an EXPERT before he has an OPINION. He’s more qualified than you to have an opinion on this, yet you're entitled enough to share yours.
@@thewiseowl8804 I've worked in ML and AI for 13 years, I think that makes me more qualified than him on this particular topic. He's literally done no AI work in his life, except some 90s classes at Berkley perhaps, so that bar is not particularly high to beat. I admit he's smarter than me, so if he spent some focused time working on actual AI, then sure...
@@martindbp A+ response. Happy to be proved wrong.
Q learning might be the next jump
Yet more nonsense from Jonathan Blow.
The argument about old jobs that are no longer relevant doesn't hold any water when the starting prompt is that these advancements are built on the backs of the prpduction of artists that did not consent to their work being used to fuel their replacements, or used at all. That is something that did not have to happen with simple economy of efficiencies that made those jobs go away, all of this production is only possible on the basis of the availability of training data.
And the "learn to code" meme is so pointless, especially when software development is probably going to be the field most affected by these developments, with millions of jobs either disappearing overnight or being reduced to massaging shitty AI prompts and output rather than doing any actual software development. Whether those jobs are valuable or not is a whole different topic, but people need to eat.
the idea that anyone should be able to sit on their copyright to this extent, like a dragon on its treasure, is massively detrimental to society
I understand very well the difficulty of monetizing creativity, but this is not the answer.
The whole idea that this is based on artists work is wrong from the start. The image generators are trained on billions of completely random junk images from the Web, not art specifically. They learn how to make images from that and what image feature go together with words. Once that initial training is done you can teach them a new style or character with a handful of examples and a few minutes of GPU time on your PC at home. If you'd remove all the artists images from the training set, nothing would change, everybody could still steal your art style simply by using an image as input instead of a word. Not much different than taking an image from one artist and telling another one "paint me something like that".
@@chrisc7265 the extent of which has been not at all?
all these ai companies just took every single piece of art every single person has ever made, gave them absolutely nothing, and now is using it to make all of them lose their jobs. the only thing you won't get the ai to generate for you is movie transcripts because movie companies have 7000 laywers that cost a billion a year. no average person is allowed to "hoard" or even own anything.
It is trained on art specifically. They view art galleries as 'datasets'. They even say so.@@grumbel45
I doubt any author of any text whose work you've read in your lifetime would consent to having their text used as training data, only to result in such an uninformed comment.
AI is NOT Art is a tool that generates images from real artists' works, photos and pictures by merging.
A Human Artist is always in process of learning for this reason there aren't mediocre artists as you said for ignorance. For instance there are professional artists and amateur artists that range from masterful to unpolished to beginner.
Do you understand?
What about the fax machine repairmen? Surely we should ban the internet, so that they fax machine repair guys can keep their jobs.
Honestly Stable Diffusion (and similar tech) are just another tool in an artist's box, if they are good at what they do. It's no different than stuff like google image search (for finding a picture of anything) or the "undo button" with drawing/tablet software, etc. Just tools that can elevate the work of good artists.
The scale and generality of usage is a million times larger than anything you used as an example.
@@BinaryDood What do you mean? I don't understand your reply.
I can't even tell if you're disagreeing with me?
The printing press made text more accessible, the uber made getting home from the bar more accessible. The wheel made horses more accessible. I'm not sure what you're getting at.
The AI models (like stable diff) produce garbage. It's useful garbage but it's still garbage. If there are low-skilled artists who can't compete with ai-gen garbage then they're in the wrong industry.
Yeah, except an undo button or image search can't fully replace whatever value you can produce as an artist. Eventually image gen will be good enough that you will not need to supervise it, so the artist will become completely obsolete.
@@SergeeeekI don't think so, because AI can't make hand-made art or paint a literal canvas. The AI has unlimited technical skill, but is limited in terms of creativity and innovation.
Artists used to fear that the printing press would be the end of their craft, or the photocopier/printer.
@@torgo_ you're right, we're left only with physical mediums for the time being. But arguably the volume of digital art vs physical art produced today leans on the side of digital. It's easier to sell, easier to share etc.
Tell them to learn to code is s sick burn.
what if low artists and mediocer artist and good artists made use of this. so they are now using what normal non artists use. with the knowledge they have accumalted and the hours they have spent trying to get good at doing art, they will match up with good artists. but the thing is that good artists will be even good. if not good then faster. so it's a win win. I don't understand why ppl are so mad about this
It's a loose-loose. You are using the system which can exist without your usage of it. Floor is so low any ignorant's work indistinguishable from a pro. There are changes than only pros can do, and in a lot of cases it is already being used for that. But everyone doing so knows it's an unavoidable free admission to be ridden off altogether.
AI art isn't human expression and brings the whole work into question.
The more a work is done by AI, the less of it is the artist's expression, the more of it is whatever the AI model found on DeviantArt.
Would the electrical power suppliers be able to keep up if AI became as competent as your average developer? A 10-15K dollar GPU has to consume so much power just to put out a subpar 200-character long output. Not to mention it's low capacity for context.
Those get more efficient every year, i don't see any reason why they would stop getting more efficient.
@@Sergeeeek because the expectation is going up. consumption is only going up. anyway, that's a huge waste of resource just to replace humans. hubris of privileged people, what can I say.
@@3_smh_3 humans aren't using 0 energy either. We just use food instead of electricity directly.
The thing is that each of those AI model deployments, while inefficient, is still serving thousands of requests simultaneously. A single person can never compete with that, even if AI sucks now.
@@Sergeeeek I'd rather not have electrical outages and carbon emissions than generate images and texts. maybe not a problem where you live. Also, AI's energy efficiency will likely never reach anyhere near human's brains. Again, hubris.
@@3_smh_3 the space is blowing up, it's a matter of time before we get hyper-specialized hardware that can only do AI stuff while being much more efficient than GPUs today. Right now most of ML stuff is still implemented in software.
What is this "How do you think human artists learn?" nonsense. He contradicts himself right off the bat. First he compares AI art using hard data as "learning like a human artist would". Then says there is NO general intelligence or knowledge behind AI at all. What a delusional thing to say lol.
I think a lot of STEM guys might unironically think the brain is a computer with the same way of handling inputs and outputs lol