The answer lives in this podcast
What is recursive self-improvement in AI, and how close is it to becoming reality?
Recursive self-improvement is the point at which AI models become capable enough to conduct the very research that makes them better — improving themselves, autonomously. Katrina Mulligan is clear: we are not there yet, but we are getting very close. The trajectory is so nonlinear and exponential that she says she cannot think five years ahead in terms of where the technology will land.
The concept is not science fiction framing. Mulligan uses it to describe a concrete threshold — one that, in her view at 3 Takeaways™, the field is approaching faster than most observers outside frontier AI development realize.
An exponential curve that makes five-year forecasts meaningless
Mulligan's position at OpenAI gives her a specific vantage point on how capability growth has actually unfolded. She points to OpenAI's scaling laws as the underlying engine: multiplying compute by 10x and data by 10x produces predictably better model performance. That predictable compounding is what makes the trajectory nonlinear.
The consequence is direct. When the rate of improvement is this steep — and when the next step could be a system that accelerates its own improvement — standard planning horizons stop working. As Mulligan explains in this episode, she cannot even think five years into the future in terms of where the technology will be. That is not hedging. It is an honest statement about the limits of forecasting when the growth curve is this compressed.
The "not quite yet" that is closer than it sounds
Mulligan's phrasing is precise: recursive self-improvement is "not quite upon us yet." That qualifier matters. It does not mean decades away. It means the gap between current capability and that threshold is now measured in the same compressed timeframes that define OpenAI's internal sense of time — where a month feels like a quarter, and three to four months at the company equates to roughly a year of experience anywhere else.
This is one of the reasons Mulligan argues that government, institutions, and enterprises need a complete recalibration of their planning units — a theme she develops further across the full conversation available on 3 Takeaways™.
"I honestly think that I am at least 30% more effective, maybe more than I was a year ago, because of how I've matured my use of these tools. I'm a better leader. I'm a better manager."
Katrina Mulligan — Head of National Security Partnerships, OpenAI.
Mulligan spent years in senior roles across the Department of Defense, the National Security Council, and the Department of Justice, including as the number two official at the Pentagon overseeing special operations. She describes her move to OpenAI as the hardest professional transition of her career — going from the most hierarchical organization on the planet to its exact opposite. That dual vantage point, inside both Washington's national security apparatus and frontier AI development, makes her testimony on AI capability trajectories unusually grounded. Her personal productivity claim — 30% more effective in a year — is offered not as a sales pitch but as a lived data point, detailed in the episode.
See also
What makes the AI revolution fundamentally different from every previous technology revolution in American history?
According to Mulligan, this is the first time in American history that a technology of this consequence is being developed exclusively by the private sector, rather than by government or the military.
What does Katrina Mulligan mean when she says the world is looking at a calendar while OpenAI is looking at a watch?
Mulligan uses this metaphor to explain that the unit of time at OpenAI is fundamentally compressed compared to anywhere else: she says internally that a month at OpenAI feels like a quarter anywhere else, meaning 3 to 4 months there equates to roughly a year of experience elsewhere.
What are Michael Pollan's three key takeaways about consciousness?
Pollan's three takeaways are that consciousness is precious, that consciousness is under siege, and that consciousness is worth defending.
Key takeaways
- Recursive self-improvement means AI becoming capable enough to drive its own research and capability gains — a threshold Mulligan places very close, not decades away.
- The growth trajectory is nonlinear and increasingly exponential, driven by OpenAI's scaling laws: 10x compute × 10x data = predictably stronger models.
- Mulligan cannot forecast five years ahead for AI — not from uncertainty, but because the curve makes such horizons genuinely unreliable.
- The proximity of this threshold is one of the strongest arguments for institutions to recalibrate how they measure time and plan around AI — a point Mulligan argues across the full episode.
This question was answered by Katrina Mulligan in episode #315 of 3 Takeaways™. The full conversation covers AI's national security stakes, the U.S.–China lead, and why government must rethink its relationship with time.
Listen to the episode on Listenly