This video is a bit unusual in style for Posy, especially visually. He's clearly grappling with what widely-available generative AI means for creators and the creative process in general, and...
This video is a bit unusual in style for Posy, especially visually. He's clearly grappling with what widely-available generative AI means for creators and the creative process in general, and where he seems to land is "as an artist, I can use this, but I should be clear that I'm doing so".
The video expects (really, invites) controversy and discussion, but YouTube comments being what they are, I thought I'd get Tildes' take on it instead.
My take is that they seem to make a clear distinction between something that is entirely AI generated and something that uses AI as samples. The latter, I think, is still a very valid use case...
I thought I'd get Tildes' take on it instead.
My take is that they seem to make a clear distinction between something that is entirely AI generated and something that uses AI as samples. The latter, I think, is still a very valid use case from a "creative" point of view as it is very similar to using samples.
Having said that, for video, I don't really agree with them claiming they are already indistinguishable from real video. We might arrive at that point in a little while, but in most cases it is often still very clear to me when visual material is AI generated. In this video it is really obvious, but that might be a choice they made when editing.
But also here, I think that if you use a bit of AI created visual material somewhere in editing a video I am mostly okay with it. In a similar sense as with samples I don't think it is all that different from using stock materials. Assuming the material used for training was actually used with consent from the people that created it.
The AI icon they created is an interesting choice, but a positive move I think.
I am not sure if I do agree with their ethical take and justification near the end. That to me seems like brushing over the thing a bit too easily.
I think YouTube commenters completely overreacted personally. He put his own twist on it, tried it out. He did some visually clever things, but the main problem is that it feels generic, because...
I think YouTube commenters completely overreacted personally. He put his own twist on it, tried it out.
He did some visually clever things, but the main problem is that it feels generic, because the AI output is generic. I think he could've pulled it off, just not like this.
A few scattered thoughts: I love Posy videos, he has a unique style and voice that make any subject fascinating to watch. It’s become clear to me that generative AI — and I have experience with...
A few scattered thoughts:
I love Posy videos, he has a unique style and voice that make any subject fascinating to watch.
It’s become clear to me that generative AI — and I have experience with text, image, and audio but not video, yet — is a tool for creatives. Sure, amateurs can use it too, but to really get the best results requires skilled manual guidance of the tech. That means understanding the parameters, carefully crafting prompts, liberal use of inpainting, postprocessing, etc. People talk about how AI models pump out finished masterpieces in seconds, but that’s just not true. In the hands of a skilled artist making creative decisions and steering the output toward a desired goal, AI can produce incredible art. But it’s not automatic. I think before long this will be widely understood, but we’re still in the novelty phase where most people haven’t yet learned the telltale signs of lazy generation.
The “AI icon” is a cute idea but opt-in watermarks are only going to be adopted by those acting in good faith. I don’t imagine this gaining much traction but it would actually be bad if it did; if people commonly started to look for the icon to indicate AI content, that’s only going to make them more susceptible to believing fakeries that don’t have it. The bottom line is, you can add metadata to generated art, and there are more complicated cryptographic ways to sign authentic content that wasn’t generated by AI… but the vast majority of artwork is (and will always be) impossible to prove the provenance of.
Piggybacking on that point, the only solution I see is becoming skeptical of everything by default. Don’t trust anything you see online. There will be reputable sources known for their policies on this, but in general we can no longer believe our eyes and ears. It sucks but there’s no putting the genie back in the bottle, however much people wish there was. Our media landscape is going to be confusing and surreal for the next decade or two.
I don’t want to rehash the debate about copyright and model training, but here’s my position on it:
a. Before the rise of AI, I was against IP as a general concept. Patents, trademarks, copyrights, you name it. I understand why we have them but I think the world would be better if there were never any legal protections for them. Of course we’ve had them for a long time now, and much of our economy is built on them, so this point sounds anachronistic and silly today.
b. Information wants to be free. The net interprets censorship as damage, and routes around it. We have a miraculous post-scarcity technology that enables instant duplication and sharing of data around the planet, and yet untold billions of dollars have been spent trying to curtail exactly that. I believe if content exists online, it should be fair play for users to do as they wish with it, be it reading once and immediately closing the tab, or archiving it, or indexing it for search, or mirroring it, or remixing it, or training a model with it.
c. Regardless of people’s positions on whether certain content “should” be used for training, that genie’s out of the bottle too. It’s too late to reverse course. At this point any further arguments are inconsequential coulda, woulda, shoulda. Looking forward, I don’t think there’s any chance of recovering that lost ground now.
I guess I posted it wrong somehow? The post preview doesn't really make clear that it's a YouTube link to Posy's channel. Is there some way to fix that?
I guess I posted it wrong somehow? The post preview doesn't really make clear that it's a YouTube link to Posy's channel. Is there some way to fix that?
Oh yeah, that's a shame. I really don't think it's as big a deal as its made out to be. I quite liked the change of the title and leaving it up. But I guess it got a bit much.
Oh yeah, that's a shame. I really don't think it's as big a deal as its made out to be. I quite liked the change of the title and leaving it up.
This video is a bit unusual in style for Posy, especially visually. He's clearly grappling with what widely-available generative AI means for creators and the creative process in general, and where he seems to land is "as an artist, I can use this, but I should be clear that I'm doing so".
The video expects (really, invites) controversy and discussion, but YouTube comments being what they are, I thought I'd get Tildes' take on it instead.
My take is that they seem to make a clear distinction between something that is entirely AI generated and something that uses AI as samples. The latter, I think, is still a very valid use case from a "creative" point of view as it is very similar to using samples.
To me the video is fairly interesting as they approach generative AI for sound similarly as to how I use LLMs, see, for example, this previous comment I made about it.
Having said that, for video, I don't really agree with them claiming they are already indistinguishable from real video. We might arrive at that point in a little while, but in most cases it is often still very clear to me when visual material is AI generated. In this video it is really obvious, but that might be a choice they made when editing.
But also here, I think that if you use a bit of AI created visual material somewhere in editing a video I am mostly okay with it. In a similar sense as with samples I don't think it is all that different from using stock materials. Assuming the material used for training was actually used with consent from the people that created it.
The AI icon they created is an interesting choice, but a positive move I think.
I am not sure if I do agree with their ethical take and justification near the end. That to me seems like brushing over the thing a bit too easily.
I think YouTube commenters completely overreacted personally. He put his own twist on it, tried it out.
He did some visually clever things, but the main problem is that it feels generic, because the AI output is generic. I think he could've pulled it off, just not like this.
A few scattered thoughts:
I love Posy videos, he has a unique style and voice that make any subject fascinating to watch.
It’s become clear to me that generative AI — and I have experience with text, image, and audio but not video, yet — is a tool for creatives. Sure, amateurs can use it too, but to really get the best results requires skilled manual guidance of the tech. That means understanding the parameters, carefully crafting prompts, liberal use of inpainting, postprocessing, etc. People talk about how AI models pump out finished masterpieces in seconds, but that’s just not true. In the hands of a skilled artist making creative decisions and steering the output toward a desired goal, AI can produce incredible art. But it’s not automatic. I think before long this will be widely understood, but we’re still in the novelty phase where most people haven’t yet learned the telltale signs of lazy generation.
The “AI icon” is a cute idea but opt-in watermarks are only going to be adopted by those acting in good faith. I don’t imagine this gaining much traction but it would actually be bad if it did; if people commonly started to look for the icon to indicate AI content, that’s only going to make them more susceptible to believing fakeries that don’t have it. The bottom line is, you can add metadata to generated art, and there are more complicated cryptographic ways to sign authentic content that wasn’t generated by AI… but the vast majority of artwork is (and will always be) impossible to prove the provenance of.
Piggybacking on that point, the only solution I see is becoming skeptical of everything by default. Don’t trust anything you see online. There will be reputable sources known for their policies on this, but in general we can no longer believe our eyes and ears. It sucks but there’s no putting the genie back in the bottle, however much people wish there was. Our media landscape is going to be confusing and surreal for the next decade or two.
I don’t want to rehash the debate about copyright and model training, but here’s my position on it:
a. Before the rise of AI, I was against IP as a general concept. Patents, trademarks, copyrights, you name it. I understand why we have them but I think the world would be better if there were never any legal protections for them. Of course we’ve had them for a long time now, and much of our economy is built on them, so this point sounds anachronistic and silly today.
b. Information wants to be free. The net interprets censorship as damage, and routes around it. We have a miraculous post-scarcity technology that enables instant duplication and sharing of data around the planet, and yet untold billions of dollars have been spent trying to curtail exactly that. I believe if content exists online, it should be fair play for users to do as they wish with it, be it reading once and immediately closing the tab, or archiving it, or indexing it for search, or mirroring it, or remixing it, or training a model with it.
c. Regardless of people’s positions on whether certain content “should” be used for training, that genie’s out of the bottle too. It’s too late to reverse course. At this point any further arguments are inconsequential coulda, woulda, shoulda. Looking forward, I don’t think there’s any chance of recovering that lost ground now.
I guess I posted it wrong somehow? The post preview doesn't really make clear that it's a YouTube link to Posy's channel. Is there some way to fix that?
That's how all YouTube links posted in ~creative are formatted.
Interesting, the video is marked private now. I was able to watch it earlier today! Wonder what happened there.
Oh yeah, that's a shame. I really don't think it's as big a deal as its made out to be. I quite liked the change of the title and leaving it up.
But I guess it got a bit much.
I didn’t look at the comments… were people up in arms? Seemed like a fine video to me.