The LLMs are on fire, and ‘AI’ is a reality in healthcare. OpenAI, Claude, Meta Llama, Mistral, Google (several models)

The general LLMs are really terrific at perhaps use cases they haven’t been ‘programmed’ specifically for. Developing prompts is essential.

Google has been taking a more specialized approach and in use cases involving imaging and specialized data analysis this might work better. We will see, we will see!

The current state of general AI medical diagnosis suffers from the same problems that affect human medical diagnosis: people, and LLMs get ‘stuck’ on the first or most likely diagniosis and get led down a rabbit hole. It becomes hard to get ‘unstuck’ from this line of reasoning. True differential diagnosis is the goal.

What is current AI great at? Its really great at automating tedious tasks, acting as a medical assistant in writing notes, automating coding, queuing tasks such as pre-cert, billing, reports, organizing raw conversations. The current state of AI is on FHIR!