How do you debug you Large Language Model Application?

Question

When ChatGPT or any other LLM start acting weird how do you debug the issues?? How often do you have to deal with such issues?? Are LLM’s unreliable for your use case?

Rajesh Kandaswamy · Accepted Answer

@sagarpatel10 Sure, happy to connect. We use LLMs for nutrition info and analysis of food diaries to come up with advice.

Rajesh Kandaswamy · Answer

This is a big challenge. We have tried to use specific instructions and structure to make this better.  But it is still a work in progress. We may put additional controls to check the output or to do a couple of queries.  But those consume a lot of resources. if you have any techniques, I’m all ears.

Sagar Patel · Answer

@rajeshkan72 what is your particular use case? The way I was thinking about tackling this by using templated responses. Also is it possible we can connect on LinkedIn or something? I’d like to ask you more questions about this issue you are having.

Sagar Patel · Answer

@rajeshkan72 Awesome, just sent you a request on LinkedIn! So it’s a personalized nutrition advice right? Looks awesome!

How do you debug you Large Language Model Application?

Replies