LLMs are being deployed in factories

I was a little surprised to read this today.

AI Assistants Join the Factory Floor

It’s about a Microsoft’s Factory Operations agent being used in an actual factory to track down causes of defects. The agent basically lives on top of the Microsoft Fabric data layer to answer user queries.

LLMs are good at this. And of course there is scope for hallucinations in this.

Here’s the thing there’s a different between using LLMs to write code or essays and using it to pseudo control stuff in an actual factory. Bad code seems abstracted from the real life.

People of course get used to the LLMs very quickly, which leads to lesser skilled software developers. In this scenario though, it’s possible that after trusting the LLM for certain things, you don’t recheck it as often, and that is a recipe for disaster further along the way.

None of these are new findings or worries about LLMs. The guess work is a feature of the product. I guess I was not expecting to see it out in the real world so soon.

But it’s here. And it’s only going to spread further.

These systems especially in factories, etc. used to be precise in the past. I am not sure how would we change to accommodate the inherent fuzziness of the LLMs.