“Situational Awareness: The Decade Ahead” is a long essay series published in June 2024 by Leopold Aschenbrenner, a former member of OpenAI’s Superalignment team. Released as a free PDF and website, it became one of the most-discussed AI forecasts of the year because of its confident, dated central claim: that “AGI by 2027 is strikingly plausible.”
The argument is built on counting orders of magnitude. Aschenbrenner traces the jump from GPT-2 to GPT-4 over roughly four years - in his framing, from “preschooler” to “smart high-schooler” - and attributes it to three compounding forces, each contributing about half an order of magnitude per year: raw compute scaling, algorithmic efficiency gains, and what he calls “unhobbling” (turning a chatbot into an agent that can use tools, reason for longer, and act). Extrapolating the same-sized leap forward, he projects machines surpassing college graduates around 2025-2026 and true superintelligence by the end of the decade.
A second pillar is the “intelligence explosion”: once you have AGI, he argues, hundreds of millions of automated AI researchers could compress a decade of algorithmic progress into about a year, vaulting from human-level to vastly superhuman quickly. To support all this, he forecasts an “extraordinary techno-capital acceleration” toward trillion-dollar compute clusters and the power infrastructure to run them before 2030, and warns that the resulting capabilities make security, alignment, and US-China competition urgent. He predicts the effort will eventually be nationalized into a government-run “Project.”
The document is a primary source for the aggressive end of the AGI-timeline debate and is frequently cited alongside Dario Amodei’s “Machines of Loving Grace” and the “AI 2027” scenario as a defining 2024-2025 statement of the short-timeline view.