Ex-OpenAI exec: Transformer architecture is the new learning bottleneck

_arohan_ · x · 2026-08-20

Jerry Tworek, Co-Founder of Core Automation and former VP at OpenAI, discusses why scaling RL hasn't delivered AGI. Despite benchmark improvements, real work remains undone. He diagnoses the Transformer architecture itself as the bottleneck, noting that models stop learning once deployed. The interview covers model capability limits and future directions.

Original post →

More from AGI Musings

AGI Musings channel →