> it shifts training compute to inference compute No, this is the change introduced by o1, what's different with R1 is that its use of RL is fundamentally dif

> it shifts training compute to inference compute No, this is the change introduced by o1, what's different with R1 is that its use of RL is fundamentally different (and cheaper) that what OpenAI did.