Remember using the models pre-thinking? I had to build thinking into ChatGPT requests manually.

(I wonder if they stole that idea from my logs…)