Not every AI prompt deserves multiple seconds of thinking: how Meta is teaching models to prioritize

Reasoning models like OpenAI o1 and DeepSeek-R1 have a problem: They overthink. Ask them a simple question such as “What is 1+1?” and they will think for several seconds before answering.

Ideally, like humans, AI models should be able to tell when to give a direct answer and when to spend extra time and resources to reason before responding. A new…