How slower does takeoff go with 10× less compute?

According to LessWrong, a ten-times reduction in AI R and D compute would slow A-G-I development by roughly six times — with a confidence interval from three-point-five to eight times. The finding comes from a mathematical model tracking how progress depends on training system performance and software efficiency. The key insight: compute cuts slow capability gains immediately, but the slowdown compounds over time, because algorithmic breakthroughs still accumulate at reduced spending rates.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton