I don’t really use advanced voice mode with ChatGPT but this isn’t completely surprising. I’ve seen similar aggressive flagging of o1s responses.
It’s not anything in your content it’s their safety model. They told me that for o1 it still needs a lot more tuning and I’m assuming same issue with Advanced Voice Mode. This is actually a really tricky problem
It’s tricky because you’re not going to notice if it lets through a response it shouldn’t have but you will immediately notice if it blocks a response it shouldn’t have. It’s a no-win scenario for the safety model as it can only ever be perceived of as having flaws.