Happy to see "inference time compute" term being used primarily nowadays - it's a much more precise and appropriate term compared to the unwieldy "test-time compute" that openai used to call it when they thought they "invented" scaling inference time.