01
Sebastian Raschka · 2025 · Ahead of AI
Why?
The four ways to get reasoning out of a model, laid out side by side: inference-time scaling, pure RL, SFT then RL, and distillation. Read it before any R1-style paper, because it tells you which of the four a given paper is actually doing. His verdict, that SFT + RL is the practical route, is the one to argue with.