Dwarkesh Patel·Apr 29, 2026

How GPT, Claude, and Gemini are actually trained and served – Reiner Pope

Name: How GPT, Claude, and Gemini are actually trained and served – Reiner Pope
Uploaded: 2026-04-29T17:20:27+00:00
Description: The talk explains how large language models (LLMs) are trained and served by detailing the impact of batch size on latency and cost, and the infrastructure limitations of GPU clusters. It highlights t