10^24 flops is a laughably small limit, below a number of models that already exist. I also don’t see how you enforce a lifetime training limit without essentially ending open source models.Absurd…

Posters: xlr8harder. xlr8harder

2 likes recorded by the source corpus.