View on GitHub

AI in HPE

An open-source, open-access resource on generative AI in health professions education.

Personas

Personas are an under-appreciated feature of generative AI.

However, you need to bear in mind that giving the model a persona doesn’t make it become that persona. When you tell an LLM who it is, you’re simply establishing the linguistic context it uses to choose which words to include in its response.

It doesn’t magically become the person you’ve asked it to be.

But this can be confusing because you can ask it to be a specialist radiographer and it’s response will be plausible and might make you think that it’s responses are accurate.

You might ask for investing advice from Warren Buffet and the LLM might respond as if it’s really Buffet (although I think the guardrails in place would prevent this). But the LLM clearly doesn’t have direct access to Warren Buffet’s mind, and cannot give you advice from his perspective.

An easy way to recognise this is when you tell an LLM that it is a mountain, and ask it to explain geological time from this perspective. An LLM is obviously not a mountain.

However, asking an LLM to respond as if it is an experienced nurse means that the total number of possible words it might use gets narrowed down and becomes more focused on the kinds of words that experienced nurses might use.

The personas you ask an LLM to take on don’t even need to be people. Imagine telling the model that it is ‘an overburdened hospital, working to provide services to a resource-constrained community’, or that it is a ‘ventilator, providing life-support to a terminally ill patient, during visiting hours’. This kind of creative prompt can generate different perspectives that provide an alternative context for a reflective exercise.


Additional resources