30 August 2026

Philosopher argues AI self-improvement has physical limits

• Toby Ord contends that recursive self-improvement, where an AI system improves itself in a loop, faces practical constraints despite being theoretically possible.
• The research-to-training cycle cannot become instantaneous because real-world experiments, model training runs, and chip manufacturing all require time.
• Physics and the speed of light impose hard limits on how fast any iterative improvement process can complete.

How it was covered

Exponential ViewAzeem Azhar

Toby Ord argues that recursive self-improvement in AI systems is mathematically possible but practically constrained by generation time, physics, and the speed of light. The newsletter agrees, noting that the full research-to-training cycle cannot shrink to zero because real-world experiments, training runs, and chip manufacturing take time.