Elon Musk
@elonmusk
Replying to @jam3scampbell
Not that it’s much consolation, but the energy efficiency of LLMs using auto-regressive transformers is extremely bad, not just in training, but also inference.
I think it’s off by several orders of magnitude.
04:42 PM · August 26, 2023 · 69.9K views
82
34
829