Google DeepMind论文：模拟对话强化学习提升精神科问诊能力
New Google DeepMind paper shows that training a model with RL on simulated patie…
将RL应用于垂直领域医疗场景的硬核实证研究，数据详实且直接验证了Agent在高风险交互中的可靠性边界，对做医疗Agent的团队极具参考价值。
New Google DeepMind paper shows that training a model with RL on simulated patients made it a better psychiatric intake interviewer, as judged by psychiatrists.
Google DeepMind 的新论文表明，通过在模拟患者身上进行强化学习（RL）训练模型，使其成为更优秀的精神科初诊访谈者，这一结论得到了精神科医生的认可。
training on simulated psychiatric intakes cut missed escalations of acute risk from 30.5% to 12.9% in adversarial cases.
在模拟精神科初诊场景下的训练，使对抗性案例中急性风险升级的漏报率从 30.5% 降至 12.9%。
In psychiatry, the intake interview is a main diagnostic tool, and key facts often surface only through trust and careful questions. Most conversational medical AI evaluations have left psychiatric cases out.
在精神病学领域，初诊访谈是主要的诊断工具，关键事实往往只有通过建立信任和提出细致的问题才能浮现。大多数对话式医疗 AI 评估都未涵盖精神病学案例。
They trained Gemini 3.1 Flash-Lite with multi-turn RL against simulated patients who reveal key facts only after rapport and targeted questions. Then 32 psychiatrists role-played patients, while 15 others graded the transcripts blind.
他们使用多轮强化学习对 Gemini 3.1 Flash-Lite 进行了训练，对手是那些只有在建立融洽关系和针对性提问后才会透露关键信息的模拟患者。随后，32 名精神科医生扮演患者角色，另外 15 人则对转录文本进行盲评。
The trained agent beat its identically prompted base model in 76.7% of 60 matched case pairs. On 256 adversarial cases, it failed to escalate acute risks 12.9% of the time, down from 30.5%.
经过训练的代理模型在 60 对匹配案例中，有 76.7% 的表现优于具有相同提示词的基础模型。在 256 个对抗性案例中，它未能升级急性风险的频率为 12.9%，低于之前的 30.5%。
If you build health agents, train them on simulated conversations, but keep clinicians supervising, since about 1 in 8 adversarial crises went unescalated.
如果您正在构建健康领域的智能体，请在模拟对话中进行训练，但务必让临床医生进行监督，因为约有八分之一的对抗性危机未被及时升级处理。
– arxiv. org/abs/2610.40476
– arxiv.org/abs/2610.40476
Title: "Training language models in simulation for psychiatric intake"
标题：《在模拟环境中训练语言模型以进行精神科初诊》
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力