- LLM without CoT is for query tasks, simple classification tasks (like Sentiment analysis and Topic labeling) and translation tasks. This kind of LLM consumes generates fewer intermediate token thus saving cost and computing resources and faster speed.
- LLM with CoT is for Math and logic tasks that rely on reasoning trace before coming up with a final answer.
Example problem:
Roger has 5 tennis balls.
He buys 2 cans of tennis balls.
Each can has 3 tennis balls.
How many tennis balls does Roger have now?
Without Chain-of-Thought (Direct Answer), the model may incorrectly compute:
