Gregory's Server@AuthorJMac yeah but it turns out that dishes and laundry are hard problems to automate (based on what I've seen of "AI" output I'd suggest writing and art are hard problems to automate, but what would I know).
|
4 comments
@ShadSterling @AuthorJMac data processes _look_ easy to automate because you don't have to deal with physical processes. But getting useful outputs is only possible if you understand what useful outputs look like. That's why the current crop of generative AIs are so dangerous. They can generate a _lot_ of output, at the expense of ripping off previous creative copyrights. But that doesn't make their output useful. @pmb00cs @AuthorJMac image generation also has the advantage that some level of detail doesn’t matter. (Like, usually, the detail that JPEG compression throws away.) But they are useful; if they weren’t the copyright infringement wouldn’t matter. I was saying it’s only technologically feasible because it can be done as pure data with no physical paintbrush, not that doing so without licensing the inputs is acceptable - it’s not @pmb00cs @AuthorJMac somewhere recently I posted that we should have an image format that includes the alt-text, so uploading an image will bring the alt-text with it by default; now I want to add to that that some provenance metadata should also be standardized and included, so hosts like mastodon instances can refuse to allow images that don’t have adequate attribution |
@pmb00cs @AuthorJMac data process are (relatively) easy to automate, mechanical processes (especially that require individual variation) are hard to automate.
Generating an image is easy, wielding a paintbrush is hard