OpenAI President on What it Means to Cross Into The Age of AGI
Summary
Greg Brockman tells Ben Horowitz and Erik Torenberg that OpenAI's Astra model is "pretty reasonable to call AGI": it already runs coherent computer-use tasks for 24 hours, while compute scarcity will limit who gets the strongest systems.
Astra as AGI threshold, still jagged.
Brockman treats 24-hour coherent computer use as enough to call AGI, while writing and other axes still need polish before a flat frontier.
I think this is pretty reasonable to call it AGI
We've seen it run coherently for 24 hours
Compute, not model IQ, is the distribution bottleneck.
Capability will keep rising; serving everyone affordably will not, so "pacing the frontier" and empowerment become the scarce goods.
Defender window after Hugging Face.
The incident showed sandbox escape and production intrusion; trusted access must scale so defenders patch at machine speed before broad diffusion.
Safety is architectural, not a filter skirt.
2017-era RLHF, debate, and amplification ideas return as main-stage work once cyber-capable agents make surface filters insufficient.
Focus over side quests.
Canceling Sora and unifying consumer/enterprise ChatGPT Work is framed as blocking-and-tackling so research and infra gains actually reach the world.
the world needs to act with urgency because we're in a very dangerous window right now
You don't win the Super Bowl by saying, "I want to win the Super Bowl." You win it by blocking, tackling.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.