https://thebullshitmachines.com/lesson-16-the-first-step-fal...
This doesn't seem to really address synthetic data, let alone RL-based reasoning.
Why would it? Once those are introduced, advancement leaves behind pure scaling.