GPT Live Questions

Agents
1

Hi everyone. Is anyone using GPT Live recently?

I migrated some of my traffic to the new S2S model and really like it but have had some edge cases. I had these in cascaded too but was able to work around it.

  1. Passing date and time between client and delegator. Saw a time where we booked an appoitnemtn for a wrong date even though user told client date. Best way for this?
  2. End call. I have the end call tool but did not delegate.
2

Would just say you have to be more precise with the prompt, atleast that’s how I fixed those issues. But I don’t use GPT Live since the voices suck compared to Cartesia Sonic 3.6. What voice are you using?

3

well ya its a totally different prompting paradigm thats why i asked. I am using gleam voice. honestly my philosophy is i could care less about quality of voice. Not trying to trick anyone if its a ai or human

4

Have an agent go through it using your existing working system prompt and then transform it into 2 separate prompts for the frontend/delegate agents using Prompting GPT-Live | OpenAI API . Explain what LLM model you currently use and then explain the issues you have with your current GPT-Live prompts (give it those aswell) and it should make a new one that’s better and fixes everything. This is what I did, and I got a result that is at parity or better than my cascaded pipeline.

And I get the perspective of not looking to trick anyone, but the voice itself makes the AI feel more capable and trustworthy to regular people, at least in my experience.