Something beautiful that generative text models will never allow again
medium.com
medium.com
I despise and dispute this fatalistic assertion! Though its not just you, I read the same thing in the NYT and other big outlets.
Stable Diffusion can already be trained on my 2020 14" laptop, and trained very quickly! LLM inference on regular computers is coming, and embeddings/finetuning will undoubtedly follow.
And if your computer is old, free peer to peer efforts like Stable Horde are already operating.
Unlike web, social media and such, generative AI is not inherently bound to corporate clouds. Maybe if more journalists pointed this out, we would escape their clutches.
Provenance matters for lots of things, including art and magic weapons. It's not just a +2 zweihander; it has a backstory, a past.
Maybe we will value the source of things more as a bulwark against the content flood. Conscious, deliberate thinking is so slow, though. We need meaningful friction without losing rationality.
Or those of us who like certain eras of music, like the 80s, draw a line in the sand and hermit ourselves in a prior age, and only its derivatives, handcrafted by a curated few, are--hesitantly--explored.
Starting from here I shall only learn from: these books priced in a handful of dollars, a collection of first principles--and curses to any future synthesis or work otherwise; now I shall only Make from those, and ignore all development hereafter. (It is kind of a burying in the sand.)