iamtrask clashes with Tom Davidson: is algorithmic progress speeding up or slowing down?

iamtrask · x · 2026-11-10

iamtrask and Tom Davidson debate whether the intelligence-explosion curve has started to bend.

Davidson argues the curve depends on growth in development-compute spend and algorithmic progress; since algorithmic progress has clearly sped up, the curve must already be bending, offering a monitoring signal for an explosion.
iamtrask counters that algorithmic progress is unavoidably about sample complexity, and the data shows it dramatically slowing.
He adds frontier labs are prioritizing scaling existing algorithms and cleaner data over dramatically more cost-efficient algorithms, because short market-share timelines, cheap capital, and closed publication practices discourage it.

More from AGI Musings