Post #3025392
2026-05-20 23:33 UTC
@JigenD@mastodon.social @SteveClough@metalhead.club @soph@grrl.me The "interesting" question about that is where new models should come from - DeepSeek only works in a world where they can fish RLHF-optimized queries from foundation models like ChatGPT. If they go away, basically the model landscape will freeze for a while and finetuning quantized models will *also* not be feasible.
This is cold turkey for coding because as libraries and structures evolve, models must always keep up - which they can't in that case.
Replies (1)
-
@JigenD@mastodon.social 2026-05-21 17:29
@ftranschel@norden.social @SteveClough@metalhead.club @soph@grrl.me That's American exceptionalism (in that it is spread by American propagandists and regulatory capture efforts, and doesn't reflect reality). China has proven they don't need anything of the sort. Just look at Seedance 2, far above OpenAI and Google's capabilities, and it's months older than Google's new pathetic Omni video model.