Chollet argues science is recursively self-improving yet yields linear progress, so AI RSI will not cause an intelligence explosion

François Chollet argues that science is a recursively self-improving system whose inputs grow exponentially while its impact grows only linearly.

He cites research headcount doubling roughly every 15 years, compute for science doubling about every 2 years, and life expectancy rising about 3 months per year since 1840. From this he concludes that AI recursive self-improvement is already happening but will not produce an intelligence explosion. This is Chollet's claim as stated in his post; the report contains no independent verification of the figures or the inference.