2026-09-08 07:00 UTC
@troed@swecyb.com @zz_james@mastodon.social @eff@mastodon.social
Firstly, I'm not taking so the assumptions in that paper at face value, but even if I did, do you know how many models were trained last year? Do you know how many were started and didn't complete their training? Do you know whether people were attempting incremental training?
*If* you accept the stated values as being correct for that particular model you can't know if that's representative of all models.
The authors of that paper have put together an interesting thought experiment, but I want some actual evidence before I accept their conclusions.
Replies (2)
-
@mu@mastodon.nz 2026-09-08 07:04
@troed@swecyb.com @zz_james@mastodon.social @eff@mastodon.social in other words, a guess. Like I said.
-
@troed@swecyb.com 2026-09-08 07:04
@mu@mastodon.nz You're free to "do your own research". As all anti-AI fanatics you're following your antivaxxer friends closely in that respect. The fact is that there exists ethically trained LLMs on properly licensed data, trained in datacenters that run on 100% renewable energy and not using potable water for cooling. Yet, you will not use such a model because your anti-AI fanaticism has nothing to do that - it's all just feels. https://www.askdifference.com/estimate-vs-guess/ @zz_james@mastodon.social @eff@mastodon.social