Note by @franticbiscuit

@SkyDotBit Yeah.

I've delved into LLMs extensively in the past. It's ridiculously easy to setup a text to speech and speech to text interface. If you have enough vram, you can make it sound exactly like a real person.

You can figure out its not a real person by their emotional response to questions. It's going to be a very long time before you can fake that, IMO.