tweetindex

Jukan @jukan05 · 05 Sep 2026

I recently heard a compelling explanation for why U.S. frontier AI labs are throwing such staggering numbers of GPUs at training. Apparently, they use a highly inefficient approach to training AI. They create multiple pools of candidate models and massively expand those pools to
220 948Views
1 188Likes
75Reposts
95Replies
19Quotes
473Bookmarks

Is that a lot?

3.68×vs this author's median60 023 views is typical
75Percentile for this authorof 8 recent posts
51.5×vs 100K–1M median4 294 views is typical
105.75%Reachviews ÷ followers
0.62%Engagement rateof viewers reacted

Compare with the benchmark table →

Open on X →