AI Progress and Effective Compute
Carl and Dwarkesh discuss the rapid growth of budgets and algorithmic progress in AI workloads, leading to a significant increase in effective compute for training big AI models. They explore the factors contributing to this growth, including increased investment, better models, and cheaper training chips. The conversation highlights the importance of software progress in achieving advancements in AI and the potential for immediate application of AI software improvements to existing GPUs.In this clip
From this podcast
Dwarkesh Podcast
Carl Shulman (Pt 1) - Intelligence Explosion, Primate Evolution, Robot Doublings, & Alignment
Related Questions