Ex-OpenAI research VP Jerry Tworek: his RL idea stalled for two years until one sentence unlocked it

cen6wkf · reddit · 2026-09-08

Jerry Tworek (ex-VP of Research at OpenAI, now CEO of Core Automation) recounts how his idea of applying RL to GPT-3 sat unfunded for two years — until his chief scientist looked at what he'd built and said "we have the GPUs now, try to scale it." His takeaway: the bottleneck wasn't skill or proof, but someone with authority finally saying go.

Original post →

More from AGI Musings

AGI Musings channel →