原推:(And, of course, the actual audio data encodes substantially more information than the tokenized text.
AI-systems trained on written text will be much more Spock-like than AI-systems that have embedded the emotional information captured by tone.)
https://twitter.com/wintonARK/status/1521230637675335683