🔮 Unbounded self-improvement and its limits #599
Plus: The deceptive popularity of open-weight models & DIY compute.
Good morning! The limits to recursion are vast. The conditions for AI systems to undergo recursive self-improvement (RSI) are exceedingly challenging. One major issue is the generation time: how long it takes for an AI system to complete a single loop of improvement. Toby Ord, a philosopher, argues that while RSI is mathematically possible, the circumstances required are practically unattainable.
The time needed for research, training, and development inherently prevents generation time from reaching zero. Real-life constraints like experiments, training runs, and chip manufacturing all take time. Even if it feels fast, the AI's progress would not accelerate infinitely. Other physical boundaries, such as the speed of light, the Bekenstein bound, and Landauer's principle, also place limits on the system's capabilities.
Unbounded recursive self-improvement is therefore not feasible. You can read Toby Ord's paper for more details. Premium members can access a plain English interactive version.
Written by urgent.news from Exponential View's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
