Tampa (Fla.) General Hospital is fine-tuning an Nvidia AI model to screen large language model traffic against its own risk and privacy standards, adding a locally run compliance layer as the health system expands its use of AI.
Palantir Technologies described the project in an Oct. 21 announcement. The approach keeps sensitive data and model processing local, on top of the safeguards Tampa General already has in place.
According to Palantir, existing tools can flag AI traffic but lack the fine-grained controls the health system wants to set for itself. To build those controls, Tampa General fine-tuned and post-trained Nvidia’s Nemotron open model through Palantir’s partnership with the chipmaker. The work improved model performance by 7% and cut latency by a factor of 22, Palantir said.
AI innovation “only works if our patients and team members can trust it,” said Jaimie Weber, MD, associate chief medical informatics and AI strategy officer at Tampa General. Keeping data local and setting its own controls will let the health system expand AI that improves patient outcomes and eases the work of staff caring for them, Dr. Weber said.
Tampa General has been a Palantir customer since 2021. The nonprofit academic health system uses Palantir’s Foundry, Ontology and Artificial Intelligence Platform programs to reduce patient hold times, improve nurse staffing ratios, respond to natural disasters and flag early signs of sepsis.
The post Tampa General trains Nvidia model to police its own AI traffic appeared first on Becker's Hospital Review | Healthcare News & Analysis.