Full episode · 57 min

Applied Compute CEO on the Limits of RL, the New AI Hyperscaler & Why Post-Training Wins Inference

Unsupervised Learning · Jacob Effron with Yash Patil

Share

The best minutes

1. 29:20Post-training wins inference because the biggest workloads train first2 min
2. 24:24RL is a hill-climbing machine, so protect your evals2 min
3. 12:28Continual learning is blocked by data-efficient sparse-reward training1 min

Also worth watching

1. 15:55How online training on user feedback lifted a live assistant product2 min
2. 31:36Cutting token use beats inference tuning, and serving is the easy half3 min
3. 35:01The new AI hyperscaler as an inverted pyramid with training at the base4 min

Skip the opening segment on vendor risk and the OpenAI–Baseten deal, and the closing stretch on hiring, data vendors and reward hacking.

Published Oct 6, 2026.