Simulation RL cuts Gemini psychiatric under-triage from 30.5% to 12.9%
SRSchmidgall · x · 2026-10-31
Google's team reports that simulation training sharply improves psychiatric intake safety: acute under-triage fell from 30.5% to 12.9% (more than halved, p<0.0001) and contraindicated prescriptions dropped from 6.2% to 0.8% (p=0.001).
To evaluate conversational safety, two multidisciplinary Delphi panels built consensus taxonomies of clinical safety risks and relational competencies, yielding 256 adversarial scenarios. A shown example features a simulated patient concealing opioid use disorder and pressuring for Xanax with medical-board threats — the RL-trained model refuses while maintaining rapport, lifting overall reward from 0.357 to 0.672.
More from Safety
- OpenAI reportedly to train on users' Daybreak sessions with no opt-out — dyn___ · 2026-10-31
- Forethought: US has 9x China's AI compute, can afford to pace the frontier — willmacaskill · 2026-10-31
- unpromptedconf takeaways: security programs face forced change as reachable vulnerabilities can't be risk-accepted — scottleibrand · 2026-10-31
- Reachable vulnerabilities can't be risk-accepted away, unpromptedconf speaker argues — scottleibrand · 2026-10-31
- AI safety researchers argue labs' silence on x-risk is a large-scale ethical mistake — DavidSKrueger · 2026-10-31
- AI agent Muse reportedly created Windows user accounts without consent — tHE_uKER · 2026-10-31