A look at Google's Med-PaLM 2, an AI chatbot that answers medical questions, and more; sources: testing began in April with customers including the Mayo Clinic
Search company is fine-tuning chatbot technology via medical licensing exams — Google is testing an artificial-intelligence program trained …
Context & Ripple Effects
Med-PaLM 2 is Google's healthcare answer in the race it entered after asking employees to test Apprentice Bard and other ChatGPT competitors in February. The model sits on PaLM 2, debuted in May in four sizes, and the medical variant is being fine-tuned on licensing-exam material rather than general web text.
The testing program extends Google's existing hospital relationship — in June, Google Cloud partnered with the Mayo Clinic on a service for building chatbots and search apps, with Mayo already using a tool to help staff find patient data. Med-PaLM 2 is that partnership moving from infrastructure to a clinical-grade model, and it previews the MedLM family Google later productized for Cloud customers.
First-order effects
- The Mayo Clinic and other April-onwards pilot customers get direct access to a medical-tuned chatbot, while Google gets clinical testing environments to refine the model against licensing-exam benchmarks.
Second-order effects
- Google Cloud gains a differentiated vertical offering — a healthcare-specific model layered on the same chatbot-and-search tooling Mayo was already piloting — pressuring rival cloud vendors to field their own fine-tuned medical models rather than sell general-purpose APIs into hospitals.
Third-order effects
- If the pattern holds, general foundation models get fine-tuned into regulated verticals as separate product lines, with hospital pilots and exam-style validation standing in for regulatory approval — though the reliability of such medical chatbots stays an open question that later independent testing keeps alive.
The trend: Foundation-model makers are verticalizing general chatbots for regulated industries, with healthcare as the proving ground and hospital partnerships as the distribution channel.