Now you've learned about the top 5 large language models or AI models that you can play with and actually solve real problems for in your practice. Remember, you never want to share PHI information with any of these models. You should always be a sensitive about what you share with these models, especially because we're in health care and HIPAA and PHI are real things. None of these tools came to be compliant in any of those things.
Of course, it's also not a tool to diagnose patients. Even many of our patients know that. But it's not here to replace clinical judgment either. Patient data should not be entered, and human accountability remains essential. I always say the human in the middle or the human in the loop, which is us as doctors, is super important and critical no matter which model you use.
I'll leave you with this. Clinicians who benefit most from AI will be the ones who use it and start to dabble, but don't trust it blindly. We are in health care after all. You should also use these models for what they're best at. Like we do with insurance, we also do coordination of care where we use different insurances for different procedures that we perform on patients. I love coordination of AI models. I use certain models for whatever their strengths are, and hopefully you have some takeaways from these videos that you can benefit from. Make sure that you are the human in the middle, so no matter what the output or response is, you're reviewing it. And again, be careful with patient data.
But my strongest recommendation is start to play with these models. You'll find which ones are great at what. You'll learn naturally yourself. And a great pro-tip would be to ask these models to ask you questions before you get a response from it. That way it has more context and gives you a more personalized response instead of something more generic.
Thank you, everyone.
Read more in my full column, "A Practical Guide to Choosing an AI Partner" from the May/June issue of Optometric Management, and tune in to my other summaries here:


