It's technically possible that the Sky voice/persona is half voice actress and half prosody/intonation ("performance") from SJ/"her". Clearly the ChatCGT tts system is flexible enough to add emotion/drama to the underlying voice, and that aspect must also have been trained on something.

Clearly a lot of people (including her "closest friends") find the ChatGPT demo to have been very similar to SJ/"her", which isn't to deny that the reporter was fed some (performance-wise) flat snippets from the voice actor's audition tape that sounded like flat sections of the ChatGPT demo. It'd be interesting to hear an in-depth comparison from a vocal expert, but it seems we're unlikey to get that.