Ex-OpenAI researcher lists 3 copiums on AI progress, still puts simulation theory at 50/50
AaronBergman18 · x · 2026-09-10
Aaron Bergman shares his three "copiums" for staying less pessimistic about AI progress:
- The Overton window is widening faster than he expected.
- Substantial uncertainty remains around non-verifiable, poorly-defined domains — he thinks they get solved, but it could take 4 years instead of 1.
He also says simulation theory is still plausible, maybe 50/50: "the world is extremely weird," amounting to "massive wildcard uncertainty injection." A big-picture musing on capability trajectories from an AI insider.
More from AGI Musings
- X debate: US elite discourse has obsessed over AGI for three years — jd_pressman · 2026-09-10
- An AI model is only as good as its harness — 0xsachi · 2026-09-10
- X thread argues 'sentience' isn't a natural category for advanced AI agents, citing 'If Anyone Builds It, Everyone Dies' — jessi_cata · 2026-09-10
- "I wanted a bicycle for the mind, we got DoorDash for Thinking" — round · 2026-09-10
- Daniel Susskind's new book on thriving in the AI age is out today in the UK — jessi_cata · 2026-09-10
- Debate: is Yudkowsky's alignment premise just nihilism in disguise? — irrepldjw · 2026-09-10