انضم إلى نوستر
2026-08-10 16:18:00 UTC
in reply to

Troed Sångberg on Nostr: No. This is the common mixup between LLMs (transformer architecture) and Diffusion ...

No. This is the common mixup between LLMs (transformer architecture) and Diffusion models (images/video/audio). They don't work at all the same.

If LLMs could store source data as you describe they would be the world's best compression systems, by a few orders of magnitude.