If you have something inside your head that you want to just show someone, and its not intended to be art, or remembered, then should you spend years practising to use some medium in order to create it yourself?
How many resources does that use, and opportunity cost in time spent, vs a occasional generation? I have a fair idea. Creating stuff is fun if that’s your goal. Dedicating time and resources to become an artist only to convey a bit of info occasionally, not so much.
People have been communicating ideas through stick figures and poorly drawn lines for about as long as people have been drawing.
Edit to add: perspective and shading are also extremely modern techniques, invented during the Renaissance. Every castle, every cathedral, every bridge, and everything before then would have been planned and communicated without any way to communicate a 3 dimensional shape in 2 dimensions
Does that not illustrate the point? Our methods and tech for expressing ideas have come a long way. Historically, there would have been more people (at least proportionally) who lacked the means to communicate their ideas as fully as they would have if they’d had access to the same education and resources as we do now. It’s an analogous situation that there are now technologies one can direct to (potentially) communicate an idea better than if one first had to achieve decent artistic competency to produce a similar result.
The difference is in the cost-benefit. Shading and perspective have minimal extra cost in materials to add to a drawing. Maybe your drawing takes a bit longer and uses a little more pencil lead/charcoal/ink/whatever, and the individual requires more training and practice to draw it well. That’s not a high cost to being able to communicate 3 dimensional shapes in 2d drawings
Machine learning requires a massive amount of compute upfront (basically a swarm of GPU servers at 100% for a few days or weeks, so easily multiple kilowatts per hour at the low end), it also requires a massive amount of training data (literally hundreds of terrabytes of data! This data is typically a mix of existing datasets with appropriate licensing and scraped data which hammers small community servers and creates extra cost and work to mitigate the extra unexpected traffic, etc. etc. oh also included in those training datasets is a lot of CSAM, private medical imagery, copyright infringing uploads, invasive medical and police report information, etc etc.) then still needs a large amount of compute once the model is trained to output an image (basically 100% of one GPU for 10-60 seconds, so maybe only a few watt hours per image but it adds up fast) and the resulting output is…no better than anything a human is capable of drawing or painting.
In aggregate we’re consuming hundreds of killowat hours of energy, consuming unthinkable amounts of water to attempt to remove hundreds of killowat hours of heat out of these systems. Small webpages operated by individuals are shutting down due to the cost of bots excessively crawling their sites looking for training data. Victims are being revictimized as the worst moments of their lives are sucked up into the datasets these machine learning algorithms are training on so they can remix and recreate these worst moments an infinite number of times. And to what end? All of this societal cost just to save you from drawing some stick figures and lines on paper? This cost benefit ain’t mathing out here
The only revolutionary thing about image generation is that there’s never been such a costly way to create images. Oh and maybe it’s kinda novel that humans aren’t the ones drawing it. The output isn’t anything new, just an amalgamation of existing ideas (seriously try having it generate anything that’s not already in its existing dataset in large numbers, it completely falls apart!)
In short, I’m not convinced that your AI generated images are communicating anything that you can’t communicate with a paper and pencil. And I know I and most people for that matter would much rather see a stick figure and poorly drawn lines that a human drew than whatever slop you poked out of the slop machine
I completely agree with the ethics problems you raise, but you’re not even trying to understand my point:
the resulting output is…no better than anything a human is capable of drawing or painting.
It’s no better than what competent humans can produce, whereas I was, from the beginning, talking about those who are not competent and likely never will be.
Maybe your drawing takes a bit longer
I don’t consider months or years of a person’s free time, for something they may not be passionate about, as “a bit” of time, and I doubt you do, either.
You might want to revisit your maths, there. There’s no way sitting at a computer learning Krita for months uses less electricity than a few prompts to an AI, and there’s no requirement that the AI in use has to be running remotely at one of those big, fuck-you data centres owned by Anthropic, OpenAI, or whoever.
Yes, because creating stuff is fun.
How many resources does that use, and opportunity cost in time spent, vs a occasional generation? I have a fair idea. Creating stuff is fun if that’s your goal. Dedicating time and resources to become an artist only to convey a bit of info occasionally, not so much.
People have been communicating ideas through stick figures and poorly drawn lines for about as long as people have been drawing.
Edit to add: perspective and shading are also extremely modern techniques, invented during the Renaissance. Every castle, every cathedral, every bridge, and everything before then would have been planned and communicated without any way to communicate a 3 dimensional shape in 2 dimensions
yeah but perspective and shading ruined art of cats.
Does that not illustrate the point? Our methods and tech for expressing ideas have come a long way. Historically, there would have been more people (at least proportionally) who lacked the means to communicate their ideas as fully as they would have if they’d had access to the same education and resources as we do now. It’s an analogous situation that there are now technologies one can direct to (potentially) communicate an idea better than if one first had to achieve decent artistic competency to produce a similar result.
The difference is in the cost-benefit. Shading and perspective have minimal extra cost in materials to add to a drawing. Maybe your drawing takes a bit longer and uses a little more pencil lead/charcoal/ink/whatever, and the individual requires more training and practice to draw it well. That’s not a high cost to being able to communicate 3 dimensional shapes in 2d drawings
Machine learning requires a massive amount of compute upfront (basically a swarm of GPU servers at 100% for a few days or weeks, so easily multiple kilowatts per hour at the low end), it also requires a massive amount of training data (literally hundreds of terrabytes of data! This data is typically a mix of existing datasets with appropriate licensing and scraped data which hammers small community servers and creates extra cost and work to mitigate the extra unexpected traffic, etc. etc. oh also included in those training datasets is a lot of CSAM, private medical imagery, copyright infringing uploads, invasive medical and police report information, etc etc.) then still needs a large amount of compute once the model is trained to output an image (basically 100% of one GPU for 10-60 seconds, so maybe only a few watt hours per image but it adds up fast) and the resulting output is…no better than anything a human is capable of drawing or painting.
In aggregate we’re consuming hundreds of killowat hours of energy, consuming unthinkable amounts of water to attempt to remove hundreds of killowat hours of heat out of these systems. Small webpages operated by individuals are shutting down due to the cost of bots excessively crawling their sites looking for training data. Victims are being revictimized as the worst moments of their lives are sucked up into the datasets these machine learning algorithms are training on so they can remix and recreate these worst moments an infinite number of times. And to what end? All of this societal cost just to save you from drawing some stick figures and lines on paper? This cost benefit ain’t mathing out here
The only revolutionary thing about image generation is that there’s never been such a costly way to create images. Oh and maybe it’s kinda novel that humans aren’t the ones drawing it. The output isn’t anything new, just an amalgamation of existing ideas (seriously try having it generate anything that’s not already in its existing dataset in large numbers, it completely falls apart!)
In short, I’m not convinced that your AI generated images are communicating anything that you can’t communicate with a paper and pencil. And I know I and most people for that matter would much rather see a stick figure and poorly drawn lines that a human drew than whatever slop you poked out of the slop machine
I completely agree with the ethics problems you raise, but you’re not even trying to understand my point:
It’s no better than what competent humans can produce, whereas I was, from the beginning, talking about those who are not competent and likely never will be.
I don’t consider months or years of a person’s free time, for something they may not be passionate about, as “a bit” of time, and I doubt you do, either.
Okay show me what we’re working with. What are you trying to communicate exactly that your drawing skills are letting down but AI can do?
Or you could actually read my other comments to know that I don’t feel the need to.
i mean if you want to do a cost:benefit analysis doing it yourself wins every time
You might want to revisit your maths, there. There’s no way sitting at a computer learning Krita for months uses less electricity than a few prompts to an AI, and there’s no requirement that the AI in use has to be running remotely at one of those big, fuck-you data centres owned by Anthropic, OpenAI, or whoever.