You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
“There’s been a lot of chatter in the industry about whether these smaller models, which are cheaper to run, will take over,” he said. “Does that mean that scaling laws are over, does that mean we don’t need as much compute, or that we’re over-investing in infrastructure? It does appear that scaling laws will continue to grow [compute needs] to provide extra capabilities, it’s just translated from frontier training to post-training to test-time compute.”