Frustrating Safe

I was chatting with a colleague recently about how they had been using an LLM to work on a project, and the LLM had told the colleague they had made a great strides in exploring the problem but at some point its time to admit defeat – that solving for X wasn’t possible.

This reminded me of my own back-and-forth I had recently with GPT. I was using GPT to internet research possible solution on how to fill a gap between a bearing surface and a metal lip.

I needed to make a 2mm tall spacer to lift this bearing surface higher(see photo), which I informed the LLM I was going to do.

A 3D Printer with the right print orientation and material is a great ways too solve this problem, but at some point in the conversation with GPT, the LLM turned on me and started giving me advice to not make my own part.

In its professional opinion I was to seek, “professional advice”.

This kind of inadvertent response is something I’ve noticed popping up more in my use of LLM’s for prototyping. Its all very frustrating. I have my own professional and lived experiences, and those real experiences have taught what acceptable limits of the material I am working with.

This LLM has none of that, it only has a matrix of training data and weights. Those weights are overvaluing assurance and impractical levels safety.

Can’t Stop, Wont Stop.