agentsfactbullishTask-specific fine-tuning substantially improves open-source LLMs on medical narrative generation tasks and narrows the gap with commercial models.Computation and Language29 Jul 2026http://arxiv.org/abs/2607.21570v1