Toby Ord: RL Post-Training Compute Now Matches Pretraining

Philosopher Toby Ord cites charts suggesting RL post-training compute has reached roughly 100% of pretraining compute, matching his earlier prediction. He argues for betting on the overall compute-capability relationship rather than single bottlenecks, and notes another 10x expansion will be very hard.

2026-10-14 ~ 2026-10-14 · 2 related posts

• Toby Ord: RL post-training compute now rivals pretraining, further 10x scaling gets hard — tobyordoxford · 2026-10-14
• Toby Ord says RL post-training compute neared parity with pretraining; bet on broad compute-capability trends — ben_j_todd · 2026-10-14