> at some point the entire compositor could probably use AI like DALL-E to enforce a consistent theme regardless of how the app styles it.
That seems eminently reasonable using today's technology (maybe with some acceleration). Certainly there's plenty of examples of style transference.
> The next generation, the application won't even need to really spend a lot of resources rendering. Just a very low resource sketch + state (e.g. what text is where to make sure that the AI renders it correctly) of what it wants to draw and the AI will automatically render it at the correct resolution, anti aliased, consistent etc.
Again, this doesn't seem out of the realm or need AGI since we have existence proofs of being able to take a low-value sketch and integrate it into a consistent theme. See https://ahrm.github.io/jekyll/update/2023/01/02/three-eyed-f... or Invoke's Stable Diffusion canvas https://www.youtube.com/watch?v=RwVGDGc6-3o&feature=youtu.be. Now sure, maybe applying it to a UI where you need things to be functional or there's things around text may pose challenges. Still, I don't know why this seems crazy, require AGI, nor do I see how ChatGPT applies.
I finally describe the most advanced application and literally say:
> However, I don't think the more advanced applications here are possible with today's generation of AI. It'll probably need some amount of AGI to problem-solve dynamically on the fly.
At what point do I claim that ChatGPT is AGI? In fact, I'm saying literally the opposite - we're nowhere close to AGI. It's possible we don't necessarily even need full AGI even for the most advanced use-case (i.e. automatically generating the UI), but certainly ChatGPT / DALL-E are NOT suitable for that, nor is it clear to me that they're in the right evolutionary ballpark (i.e. maybe you need AI that works totally different).
> at some point the entire compositor could probably use AI like DALL-E to enforce a consistent theme regardless of how the app styles it.
That seems eminently reasonable using today's technology (maybe with some acceleration). Certainly there's plenty of examples of style transference.
> The next generation, the application won't even need to really spend a lot of resources rendering. Just a very low resource sketch + state (e.g. what text is where to make sure that the AI renders it correctly) of what it wants to draw and the AI will automatically render it at the correct resolution, anti aliased, consistent etc.
Again, this doesn't seem out of the realm or need AGI since we have existence proofs of being able to take a low-value sketch and integrate it into a consistent theme. See https://ahrm.github.io/jekyll/update/2023/01/02/three-eyed-f... or Invoke's Stable Diffusion canvas https://www.youtube.com/watch?v=RwVGDGc6-3o&feature=youtu.be. Now sure, maybe applying it to a UI where you need things to be functional or there's things around text may pose challenges. Still, I don't know why this seems crazy, require AGI, nor do I see how ChatGPT applies.
I finally describe the most advanced application and literally say:
> However, I don't think the more advanced applications here are possible with today's generation of AI. It'll probably need some amount of AGI to problem-solve dynamically on the fly.
At what point do I claim that ChatGPT is AGI? In fact, I'm saying literally the opposite - we're nowhere close to AGI. It's possible we don't necessarily even need full AGI even for the most advanced use-case (i.e. automatically generating the UI), but certainly ChatGPT / DALL-E are NOT suitable for that, nor is it clear to me that they're in the right evolutionary ballpark (i.e. maybe you need AI that works totally different).