Post

So this is a fascinating article about how a family used ChatGPT to help (emphasis on *help*, not on doing everything) diagnose the rare genetic disorder their kid had. While I know it will make some people angry, I think there are a few good lessons in here: www.bostonglobe.com/2026/09/12/m...

Both the parents had some level of expertise in both AI and medical science. They focused on primary data and primary sources (and made it quite clear to the tools that they wanted clear links to sources for them to check themselves). Importantly, they worked WITH medical professionals...

... to help interpret the results, rather than assuming that the tool replaced the professionals. They still had to deal with some false alarms (though in this case it was because of bad input data, not AI hallucinations), which is why it was useful that the parents were careful.

Does everybody know you can explicitly tell Claude not to hallucinate? I learned this and gave it these rules months ago, and it’s remarkable how different the results are! www.icloud.com/notes/090yXE... 

If you use AI as a research *assistant*, it's great. If you use it as your primary researcher, it's *terrible*.

Exactly, librarian or research assistant , not a replacement for your own thinking and judgment

Yeah this is not a story of “AI” this is a story of a higher search tool They didn’t use the machine to diagnose or tell them what was wrong they used it to find main sources that they themselves read and interpreted

And it still seems to me that if we were paying the actual cost to operate one of these LLMs, enough for it to be profitable, it would be cheaper to hire a live assistant. $1 trillion pays for a lot of research.