The demos on the Jolly Roger site are hilarious. The bot sounds so realistic, and the bot even had the audacity to ask the scammer, “Are you a real person?!”
“Hold on, there’s a bee on my arm”
“Hold on, there’s a bee on my arm”
Honestly, if you combined it with a sufficiently powerful GPU, Whisper (maybe even FasterWhisper or WhisperX due to increased speeds), and fed those into the prompts it would probably work even better and be more realistic.
[0] - https://jollyrogertelephone.com/now-gpt-powered-jolly-roger-...