Hi OpenAI Team,
I encountered an issue while using ChatGPT-4o in Voice Mode.
I clearly asked:
“What ever happened to Jenkins, the deserter who went to North Korea?”
However, the model incorrectly responded with a summary about James Joseph Dresnok, a completely different U.S. defector to North Korea.
This seems like a context attribution failure, especially considering that the transcription was accurate and explicitly said “Jenkins.”
You can see the attached screenshot showing both the transcript and the incorrect response.
Feedback:
- This kind of error undermines trust in voice-based interactions.
- It’s especially concerning since Jenkins and Dresnok are both real historical figures but with distinct stories.
- I’d expect the model to disambiguate based on the explicit mention of “Jenkins.”
Thanks!
