k.Frontier
FrontierOpenAI Blog
KeyRank 65How AI training scales
OpenAI's discovery that gradient noise scale predicts training parallelizability removes a fundamental constraint on AI scaling, suggesting batch sizes can grow far larger without diminishing returns. This reframes AI training from art to science, with direct implications for how quickly future systems can be developed.
Why it ranks · · Gradient noise scale metric predicts parallelizability across wide range of tasks · December 2018
Read full story