Asked about Parameters that change the output
How do you make a model give shorter answers?
An answer that works
Ask for it in the prompt. max_tokens is a hard cut-off that truncates mid-sentence rather than an instruction to be concise. Keep max_tokens as a safety limit and check stop_reason to detect when it fires.
Reading an answer is not being able to give one. The section that teaches this is Parameters that change the output, 2 hours of The Messages API.
Learn this properlyAlso asked about The Messages API
What are the roles in a chat request and what is each for?What is stop_reason and why check it?Does max_tokens make the model give shorter answers?Are LLM APIs stateful or stateless?Why does a long conversation get expensive faster than you would expect?How would you keep a long-running chat affordable?What does temperature do?Is temperature 0 deterministic?
All 10 questions on The Messages API