Tuesday, December 23, 2025

Why I think AI is Less Dangerous Than Everyone Assumes

1. A professor tells me a student turned in a paper with this sentence: "insert your name here".  If you are so lazy that you cannot read it first before submitting it....

2. While waiting for Sherline to respond to my request for technical assistance with my shorted out spindle motor, I decided to see if I could use AI to 3D print the refractor lens cell.  

CoPilot understood my prompt and seemed to have produced a Python script that a FreeCAD add on can convert it to a Free CAD object 

Okay.  I asked CoPilot to help me install the add on.   The first instructions failed.  So it walked me through installing git and eventually I got the add on installed in FreeCAD.  

After multiple false starts (many commands do not work as CoPilot thought and to be fair, they were logical assumptions).  I was able to get objects created.  Continued attempts to get an isometric view of the parts continued failing.  I may yet get this operational using CoPilot but it is a lot of "this is the error, what next?"  I suspect with enough hours, I will get something useful out of it. 

Like court decisions and briefs filed by lawyers, AI is so often wrong that it is at best a useful tool. It is not a replacement for a human. I asked it if it was learning from its misstatements about how FreeCAD worked.  The answer was that while it used its mistakes to guide me toward a real solution, it does not feed that knowledge back into its knowledge based.  This seems like a severe deficiency  

No comments:

Post a Comment