- Joined
- Jul 6, 2021
For real. The uncanny ML speech synthesis stuff like ElevenLabs or Suno or whatever is just so uncomfortable to listen to. I hate it.Computer speech should sound like stereotypical robots,
There's also a huge psychological impact. Have you heard of the ELIZA effect? All the way back in 1966 a researcher at MIT developed one of the first "AI chatbots" that was really just a bunch of if / then logic and canned responses.
Nonetheless, people he would recruit for the experiment would talk to it as if it was a real person and engage for hours despite the fact the program had literally no understanding of what the user was even inputting.
LLMs are this times 10x. Add in something like realistic almost-human sounding voices and you've multiplied it by 100x. Normalfags are dumb and even now with the current state of machine learning people think it's like an actual sentient program talking to them. You can't explain to them what a token is.
At least if it was something like a vocoded robot voice like from a movie, normalfags would understand that they are talking to a computer program that's fallible. People trust ChatGPT and similar far too much because they literally think it's a person and not a token predictor. Maybe that's an interesting plot device to explain why they sound like that in media set in the future?
