This just weirds me out way too much, how on earth can AI be this good already? I remember when detecting the picture of a cat was an unsolved problem in computer science. Those avocado arm chairs...
This just weirds me out way too much, how on earth can AI be this good already? I remember when detecting the picture of a cat was an unsolved problem in computer science.
Those avocado arm chairs are all brilliant. Like, ask a production design student to sketch this up as an assignment and you'd be lucky to get any of those.
Probably very, but just picking two photographs and combining them, preserving meaning, is impressive. I'll just say it, this is creativity. This is absolutely the right path to start just about...
Probably very, but just picking two photographs and combining them, preserving meaning, is impressive. I'll just say it, this is creativity. This is absolutely the right path to start just about any creative project. Going into real-world production, actually thinking about how this furniture would be made and used is another topic, but what's lacking here is training data, not skill.
The moment they figure out how to make the learning process more general, like teaching AI to productively search the internet or companies start offering carefully trained AI to do steps like turning a vague image of an object into a schematic sketch fit for production, I could actually see this technology used in real-world work that never used a computer before. It's too fuzzy an outlook to say concretely what it will be used for (I doubt we'll see novelty furniture production automated completely, any time soon) but it feels like it has potential to change things up.
I think this has a lot of potential. The reason it should work in practice (unlike many supposed GPT-3 applications) is that, if you don't like the image you get, you can just ask again. Hopefully...
I think this has a lot of potential. The reason it should work in practice (unlike many supposed GPT-3 applications) is that, if you don't like the image you get, you can just ask again. Hopefully they can get CPU usage down enough to generate several images and let you pick the best. I could see this replacing a lot of dumb clip art. (And then people getting used to and then tired of its style.)
If you think about it, that's how search engines work. The first link isn't necessarily the best one, but you want to find something good in the top few links.
As they say, "with varying degrees of reliability." The key is finding applications where reliability doesn't matter too much because someone is going to review the results.
Depending on the quality of the output, the possibilities for abuse could be pretty scary though. If Bing launches something like this, they will probably have to block some inputs.
That's going to be the hard part. Honestly, I'm having a hard time trying to figure out what that will be. Perhaps people will just use it as a way to brainstorm visual concepts? I think the...
As they say, "with varying degrees of reliability." The key is finding applications where reliability doesn't matter too much because someone is going to review the results.
That's going to be the hard part. Honestly, I'm having a hard time trying to figure out what that will be. Perhaps people will just use it as a way to brainstorm visual concepts?
I think the reliability issue is perhaps bigger than we think. If you look at their "food of Japan" example, you see that everything turns into sushi. And this good boy here is Japan's national animal. It would seem that like most AI examples, the set of images you choose to train the AI on is vitally important.
Well, the idea I had is that people without artistic talent can do a few searches and maybe find an image they like to illustrate their article. Which, admittedly, is something you could do with...
Well, the idea I had is that people without artistic talent can do a few searches and maybe find an image they like to illustrate their article.
Which, admittedly, is something you could do with Google Image search, and you could look for creative commons images to avoid copyright issues. But maybe this would be better for creating original images that aren't already out there?
So maybe the result would be a creative tool? Not world-changing, but sort of neat, anyway.
This just weirds me out way too much, how on earth can AI be this good already? I remember when detecting the picture of a cat was an unsolved problem in computer science.
Those avocado arm chairs are all brilliant. Like, ask a production design student to sketch this up as an assignment and you'd be lucky to get any of those.
It would be interesting to know what the closest chair images are the training set.
Probably very, but just picking two photographs and combining them, preserving meaning, is impressive. I'll just say it, this is creativity. This is absolutely the right path to start just about any creative project. Going into real-world production, actually thinking about how this furniture would be made and used is another topic, but what's lacking here is training data, not skill.
The moment they figure out how to make the learning process more general, like teaching AI to productively search the internet or companies start offering carefully trained AI to do steps like turning a vague image of an object into a schematic sketch fit for production, I could actually see this technology used in real-world work that never used a computer before. It's too fuzzy an outlook to say concretely what it will be used for (I doubt we'll see novelty furniture production automated completely, any time soon) but it feels like it has potential to change things up.
I think this has a lot of potential. The reason it should work in practice (unlike many supposed GPT-3 applications) is that, if you don't like the image you get, you can just ask again. Hopefully they can get CPU usage down enough to generate several images and let you pick the best. I could see this replacing a lot of dumb clip art. (And then people getting used to and then tired of its style.)
If you think about it, that's how search engines work. The first link isn't necessarily the best one, but you want to find something good in the top few links.
As they say, "with varying degrees of reliability." The key is finding applications where reliability doesn't matter too much because someone is going to review the results.
Depending on the quality of the output, the possibilities for abuse could be pretty scary though. If Bing launches something like this, they will probably have to block some inputs.
That's going to be the hard part. Honestly, I'm having a hard time trying to figure out what that will be. Perhaps people will just use it as a way to brainstorm visual concepts?
I think the reliability issue is perhaps bigger than we think. If you look at their "food of Japan" example, you see that everything turns into sushi. And this good boy here is Japan's national animal. It would seem that like most AI examples, the set of images you choose to train the AI on is vitally important.
Well, the idea I had is that people without artistic talent can do a few searches and maybe find an image they like to illustrate their article.
Which, admittedly, is something you could do with Google Image search, and you could look for creative commons images to avoid copyright issues. But maybe this would be better for creating original images that aren't already out there?
So maybe the result would be a creative tool? Not world-changing, but sort of neat, anyway.
This will kill the stock image business.
Or maybe it evolves? The algorithm needs training data.