I sometimes write posts that might seem anti-“AI,” but the truth is more complex. I am always guarded in how I evaluate statistical models like generative LLMs, because patient safety comes first. However, generative LLMs are amazing (kudos to the engineers, the researchers, and the people who built the internet and wrote things on it (which were then used to train the generative models)).

To reconcile my “AI” skepticism with my genuine amazement at generative LLMs, I have needed to think differently. I needed to stop asking what we do in medicine, and how an “AI” might fit in, and to start asking what a generative LLM really is, and how *that* might fit into our ultimate goal of improving patient care.

I describe my thoughts in more detail in this post.