Most methods that optimize LLM prompts and agent workflows assume that task-specific output schemas, extraction instructions, and evaluation criteria are pre…
机构:中科大
来源:arXiv 2609.34829 | AI4Papers 论文推荐平台