Google DeepMind researchers trained Gemini 3.1 Flash-Lite through multi-turn reinforcement learning with simulated patients to improve psychiatric intake interviews. In a randomized, double-blind crossover objective structured clinical examination, 32 psychiatrists preferred the trained model in 76.7% of paired comparisons, with a p-value below 0.0001. Missed escalations of acute safety risks fell from 30.5% to 12.9% across 256 adversarial scenarios, while the clinical rubric score across 127 psychiatrist-validated scenarios rose from 0.379 to 0.706. The findings come from a preprint and have not yet undergone formal peer review or testing with real patients.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
