Shopify CTO Mikhail Parakhin argues big-tech ML talent evaluation is flawed, pointing to Recraft outperforming larger labs with fewer GPUs
Recraft beat Microsoft and Meta using less training data.
#488kache@YACINEMTB@MParakhin ETH Zurich papers are often done on a single GPU
Mikhail Parakhin@MParakhinModel training is a game where GPUs and data are an overwhelming advantage. So when Recraft beats xAI, DeepSeek, Meta, BFL, Microsoft, etc. with a tiny fraction of the resources, the conclusion is: big-company ML talent selection is broken. Very different from "AI experts" :-)
#488kache@YACINEMTB@MParakhin Check out http://puffer.ai (pufferlib). RL training in seconds with a single consumer gpu
Mikhail Parakhin@MParakhin@yacineMTB :-) Not to compare myself to the greats, but I always run everything on one GPU - I want to compete in areas where smarts and experience are an advantage, not the raw compute power.
@yacineMTB :-) Not to compare myself to the greats, but I always run everything on one GPU - I want to compete in areas where smarts and experience are an advantage, not the raw compute power.
kache@yacineMTB@MParakhin ETH Zurich papers are often done on a single GPU