I would somewhat suggest older tech like books and flow charts over the latest LLM’s. I’m not saying LLM’s wouldn’t work. Just that I don’t really trust LLM’s, and a simpler flowchart based system won’t suddenly start talking about goblins for inscrutable reasons.
We have digital flowcharts. That’s essentially what a traditional CDSS is.
On the LLM skepticism, it’s possible they have rare hallucinations, but I don’t think it’s likely that spontaneous rambling about goblins is a meaningful concern. In my experience, you have to try pretty hard to get GPT or Claude to have mental breakdown. It’s not usually something that happens for mundane use.
The bottom line is whether, all things considered, the results for patients are better.
You also need to weigh every “goblin moment” against the human equivalent—“couldn’t be bothered” “underslept”, “whoopsie daisy”, and so on.
I do think that, among other things, you’re underestimating the difficulty most people will have using LLMs effectively for healthcare. I agree they can be extremely useful, but there are also many non-obvious failure modes where people can lead themselves astray. It’s still a pretty uncommon skillset and benefits from the user having an above-average level of background understanding and introspective awareness. I don’t think that will keep being true, but I think it’s true now.
Are we comparing the current system to LLM’s? Or a well designed digital flowchart system to LLM’s? I think the first one is a win for the LLM’s. Not sure on the second one.
I would somewhat suggest older tech like books and flow charts over the latest LLM’s. I’m not saying LLM’s wouldn’t work. Just that I don’t really trust LLM’s, and a simpler flowchart based system won’t suddenly start talking about goblins for inscrutable reasons.
We have digital flowcharts. That’s essentially what a traditional CDSS is.
On the LLM skepticism, it’s possible they have rare hallucinations, but I don’t think it’s likely that spontaneous rambling about goblins is a meaningful concern. In my experience, you have to try pretty hard to get GPT or Claude to have mental breakdown. It’s not usually something that happens for mundane use.
The bottom line is whether, all things considered, the results for patients are better.
You also need to weigh every “goblin moment” against the human equivalent—“couldn’t be bothered” “underslept”, “whoopsie daisy”, and so on.
I do think that, among other things, you’re underestimating the difficulty most people will have using LLMs effectively for healthcare. I agree they can be extremely useful, but there are also many non-obvious failure modes where people can lead themselves astray. It’s still a pretty uncommon skillset and benefits from the user having an above-average level of background understanding and introspective awareness. I don’t think that will keep being true, but I think it’s true now.
I agree, however most people have difficulty “using” PCPs effectively for healthcare as well.
Peoples’ ability to articulate their symptoms doesn’t change much by stepping into a medical office.
Whether you’re prompting an AI or talking to a human, ultimately it’s a matter of the words you say and how those words are interpreted.
Fair enough.
Are we comparing the current system to LLM’s? Or a well designed digital flowchart system to LLM’s?
I think the first one is a win for the LLM’s. Not sure on the second one.