Palace/Gallery

Leopold Aschenbrenner

aiagiopenaisuperalignmentnational-securityresearcher

Leopold Aschenbrenner

Former OpenAI researcher who worked on the Superalignment team under Ilya Sutskever. Published Situational Awareness: The Decade Ahead in June 2024, a ~165-page essay series arguing AGI arrives by ~2027 and superintelligence by ~2030, framing the development as the most consequential national security challenge since nuclear weapons.

Background

  • Worked at OpenAI on technical alignment research (weak-to-strong generalization, the study of whether a small model can align a larger one)
  • Left OpenAI shortly before publishing Situational Awareness
  • Describes himself as an "AGI Realist" -- neither a doomer nor an e/acc
  • Background in economics and progress studies (references theoretical economic models of the path to AGI)

Key Ideas

  • OOMs framework: AI progress measured in orders of magnitude of effective compute, not benchmarks. Three drivers: compute scaleup, algorithmic efficiency, unhobbling.
  • Intelligence explosion: Automated AI researchers (100M at 100x human speed) compress a decade of progress into ~1 year.
  • AGI as national security: Superintelligence provides decisive military advantage; must be treated like nuclear weapons, not consumer tech.
  • The Project: Prediction that by 2027-28, the US government will form a Manhattan Project-style AGI effort.
  • AGI Realism: Third way between doomers and e/accs -- take the power seriously, America must lead, and we must not screw it up.
  • Security as the #1 priority: AI lab security is catastrophically poor; algorithmic secrets being leaked to adversaries is the single biggest near-term threat.

Sources