If this were posted anywhere else on the internet, or by someone less trustworthy than skybrian, I would think the author was some actual artist trying to pull one over on me. Did you really think...
If this were posted anywhere else on the internet, or by someone less trustworthy than skybrian, I would think the author was some actual artist trying to pull one over on me. Did you really think some computer algorithm could turn these rough MSPaint doodles into detailed cohesive artwork? LOOOOOL
Seeing this gives me a similar feeling to the author, and to the one I had when I read the transcript of the “sentient” AI. It’s breathtaking and staggering at the same time — both for what it’s already able to do and for its potential in the future.
@Whom had a great point in the AI topic that has stuck with me ever since:
we're not that far away from when informed but nontechnical laypersons are no longer able to judge this themselves and we all have to rely on what experts tell us
I don’t know if she meant it to apply only to AI, but I’ve been thinking about it as it extends to stuff like this that we associate with human creativity. How long until my students are turning in machine-generated essays that I cannot distinguish from human writing, for example? I don’t feel like it’s that far off…
It's an article I saw it on Hacker News and I don't actually have any direct evidence that it wasn't faked, but unlike the "sentient" AI, I consider it entirely plausible. People on Hacker News...
It's an article I saw it on Hacker News and I don't actually have any direct evidence that it wasn't faked, but unlike the "sentient" AI, I consider it entirely plausible. People on Hacker News are very excited about these tools, and I can see why.
That's based on trying out the online image generation websites. The img2img tool was just released and I don't have a good graphics card so I haven't played around with it myself. There are notebooks for running it online, but I've seen hints that it might come to dreamstudio.ai soon so I'll wait.
Your concerns about essays are sort of similar to the ones computer science teachers have for this upcoming school year, due to GitHub CoPilot. (At least, for beginner programming classes.)
Semi-hijack I wonder what difference it will make if you can't distinguish the essays' origins? A student who taps an AI allows one skill to whither, but hones another, and both may be valuable....
Semi-hijack
I wonder what difference it will make if you can't distinguish the essays' origins? A student who taps an AI allows one skill to whither, but hones another, and both may be valuable. My instinct says otherwise. And my inner cynic says society has been usurped (if it was ever otherwise) by those who can and do present work generated by others (human or ai) as their own, regardless of whether it is "better" by any consensual standard.
The thing that impresses me the most about this (not that this kind of pipeline comes completely out of left field to me) is that there's this reasonable workflow to it. Ya know, for an AI model...
The thing that impresses me the most about this (not that this kind of pipeline comes completely out of left field to me) is that there's this reasonable workflow to it. Ya know, for an AI model running on a command line, this has a nice workflow. You give it a baseline image and a prompt (or presumably just a prompt would work too, if you don't care too much.) and you pick out results. You change them around, refine them. It's all quite intuitive and actually reasonable. There's no technobabble necessary, no "you need to set the gradient in the optimizer to a smaller value, or else the diffusion will be unstable and the bit shift vector will not converge" or whatever (imitating my own field, how'd I do?). A literal child could operate this. You've got strength, which is the "creative liberties", a prompt and a few technical ones no one gives a crap about. And you can get good results without knowing fuck all about what's going on under the hood. It handles like a friendly artist would. "Paint me something like this, just pretty. Take all the liberties you like. I was going for a seattle skyline btw." I'm impressed.
I fooled around with this a lot today (actually installed everything on my computer). It works nicely and can yield some incredible results. It can also produce some garbage tier nonsense if you...
I fooled around with this a lot today (actually installed everything on my computer). It works nicely and can yield some incredible results. It can also produce some garbage tier nonsense if you give it the wrong prompts - it feels like it will be the same type of skill as knowing how to look things up on google. Which words to use, how to phrase things, how to iterate on an image to get the desired results.
[…] it won’t be fresh and new for long. This thing I’m feeling is not much different from how I felt using email for the first time - “Grandma got my message already? In Florida? In seconds?” It was the nearest thing to magic my child-self had ever seen. Now email is the most boring and mundane part of my day.
There is already much talk about practical uses. Malicious uses. Downplaying. Up playing. Biases. Monetization. Democratization - which is really just monetization with a more marketable name.
I’m not trying to get into any of that here. I’m just thinking about those 4.2 gigabytes. How small it seems, in today’s terms. Such a little bundle that holds so much.
If this were posted anywhere else on the internet, or by someone less trustworthy than skybrian, I would think the author was some actual artist trying to pull one over on me. Did you really think some computer algorithm could turn these rough MSPaint doodles into detailed cohesive artwork? LOOOOOL
Seeing this gives me a similar feeling to the author, and to the one I had when I read the transcript of the “sentient” AI. It’s breathtaking and staggering at the same time — both for what it’s already able to do and for its potential in the future.
@Whom had a great point in the AI topic that has stuck with me ever since:
I don’t know if she meant it to apply only to AI, but I’ve been thinking about it as it extends to stuff like this that we associate with human creativity. How long until my students are turning in machine-generated essays that I cannot distinguish from human writing, for example? I don’t feel like it’s that far off…
It's an article I saw it on Hacker News and I don't actually have any direct evidence that it wasn't faked, but unlike the "sentient" AI, I consider it entirely plausible. People on Hacker News are very excited about these tools, and I can see why.
That's based on trying out the online image generation websites. The img2img tool was just released and I don't have a good graphics card so I haven't played around with it myself. There are notebooks for running it online, but I've seen hints that it might come to dreamstudio.ai soon so I'll wait.
Your concerns about essays are sort of similar to the ones computer science teachers have for this upcoming school year, due to GitHub CoPilot. (At least, for beginner programming classes.)
The founder says DreamStudio will have it next week.
Semi-hijack
I wonder what difference it will make if you can't distinguish the essays' origins? A student who taps an AI allows one skill to whither, but hones another, and both may be valuable. My instinct says otherwise. And my inner cynic says society has been usurped (if it was ever otherwise) by those who can and do present work generated by others (human or ai) as their own, regardless of whether it is "better" by any consensual standard.
The thing that impresses me the most about this (not that this kind of pipeline comes completely out of left field to me) is that there's this reasonable workflow to it. Ya know, for an AI model running on a command line, this has a nice workflow. You give it a baseline image and a prompt (or presumably just a prompt would work too, if you don't care too much.) and you pick out results. You change them around, refine them. It's all quite intuitive and actually reasonable. There's no technobabble necessary, no "you need to set the gradient in the optimizer to a smaller value, or else the diffusion will be unstable and the bit shift vector will not converge" or whatever (imitating my own field, how'd I do?). A literal child could operate this. You've got strength, which is the "creative liberties", a prompt and a few technical ones no one gives a crap about. And you can get good results without knowing fuck all about what's going on under the hood. It handles like a friendly artist would. "Paint me something like this, just pretty. Take all the liberties you like. I was going for a seattle skyline btw." I'm impressed.
It's what we strive for.
There's an example of integrating stable diffusion with Photoshop that's pretty cool.
I fooled around with this a lot today (actually installed everything on my computer). It works nicely and can yield some incredible results. It can also produce some garbage tier nonsense if you give it the wrong prompts - it feels like it will be the same type of skill as knowing how to look things up on google. Which words to use, how to phrase things, how to iterate on an image to get the desired results.
From the article: