While the debate rages on regarding the use cases and appropriateness of the various commercial LLM products, I did quite enjoy spending a few hours setting up my own chat instance and using langchain to embed my own content – in this case, a few pages from our Hospital HealthPathways into a transient index.
The results are indeed quite fun.
If @jon_herries needs another use case with which to evaluate for benefits and harms, this may be another interesting knowledge management tool …
Thanks Ryan. LLMs certainly open up many opportunities for new ways of interacting with clinical pathways.
The team at Streamliners is actively investigating applications of LLMs to HealthPathways. Beyond exploring the opportunities we will also be working through a range of considerations including clinical safety/hallucination risk, bias, privacy, security, and scalability that will need to be covered off before this technology can be deployed into clinical practice.
I’m very happy to hear from anyone interested in this area.
Certainly not out to show you a direction of travel for Streamliners; apologies if I’ve stepped some toes. This was primarily an experiment in coding for myself as part of the journey to learn more about how folks were leveraging LLM technology – such as in the patient-communication module for Epic, etc.
It will be interesting to see how future implementations of LLMs go about addressing guardrails and hallucination prevention; it may be part of the prompt structure or a new type of reinforcement tuning to reduce risk of straying from the intended content – or even something else entirely …
The more experiments the better - the health system certainly could do with some technological breakthroughs! Just signalling that we’re actively working in this area and keen to help get good ideas into practice.
