Again, they’re tools. Some of the most useful applications for LLMs I’ve worked on are never even seen by human eyes, like ranking, then ingesting documents and filling out json in pipelines. Or as automated testers.
Another is augmented diffusion. You can do crazy things with depth maps, areas, segmentation, mixed with hand sketching to “prompt” diffusion models without a single typed word. Or you can use them for touching up something hand painted, spot by spot.
You just need to put everything you’ve ever seen with ChatGPT and copilot and the NotebookLM YouTube spam out of your head. Banging text into a box and “prompt engineering” is not AI. Chat tuned decoder-only LLMs are just one tiny slice that a few Tech Bros turned into a pyramid scheme.
Again, they’re tools. Some of the most useful applications for LLMs I’ve worked on are never even seen by human eyes, like ranking, then ingesting documents and filling out json in pipelines. Or as automated testers.
Another is augmented diffusion. You can do crazy things with depth maps, areas, segmentation, mixed with hand sketching to “prompt” diffusion models without a single typed word. Or you can use them for touching up something hand painted, spot by spot.
You just need to put everything you’ve ever seen with ChatGPT and copilot and the NotebookLM YouTube spam out of your head. Banging text into a box and “prompt engineering” is not AI. Chat tuned decoder-only LLMs are just one tiny slice that a few Tech Bros turned into a pyramid scheme.