Prompt or Policy? Optimizing Physician Slate Presentations with LLM User Simulators
To optimize physician presentations without relying on real patient interactions, we develop an offline framework that uses persona-conditioned LLMs to simulate patient choices.
ProofPaper ↗
Key points
- Choosing the right physician is a key challenge for patients seeking medical care.
- We formulate slate presentation optimization: given a patient's medical case and a fixed, ranked list of physicians, we generate short, case-specific summaries that make doctors easier to compare, while remaining grounded in the information in their profiles.
- We compare automatic prompt-level optimization with GEPA, parameter-level optimization with GRPO, and their combination across two model sizes.
- These results show that prompt-level and parameter-level optimization play complementary roles in improving how physician recommendations are presented.
Sources (1)
- [1]Prompt or Policy? Optimizing Physician Slate Presentations with LLM User SimulatorsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 12:52 PM
To optimize physician presentations without relying on real patient interactions, we develop an offline framework that uses persona-conditioned LLMs to simulate patient choices.
Choosing the right physician is a key challenge for patients seeking medical care.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026Teach Me After You Learn It: Lookahead Self-Distillation for RLVR
- Oct 13, 2026A Few More Tokens Ahead: Reinforcing Parallelism in Diffusion LLMs
- Oct 12, 2026Meet the Anthropic Engineer Keeping Its AI Agent Trustworthy
- Oct 12, 2026unslothai/unsloth v0.1.906-beta: Browser Use + Qwen-Image-2.1-Turbo
- Oct 12, 2026Separating Prompt Harmfulness from Interaction Safety via Partial Generation
- Oct 11, 2026Every Model That Can Be Run On 10-16GB VRAM Ranked