Keep in mind that tokens are LLM units of thought; the only moment the model does any computation is when generating tokens. Therefore, asking it to be succinct means effectively dumbing it down.