Parsing emotions in vocal inflections (and reliably producing them in vocal output) seems quite under-hyped in this release.
That seems to represent an entirely new depth of understanding of human reality.
That seems to represent an entirely new depth of understanding of human reality.
Having a reliable emotional model of a person based on their voice (or voice + appearance) can be useful in a thousand ways.
Which seems to represent a new frontier.
These LLM are nothing more than really big spreadsheets.
"Um it's just a big spreadsheet" just isn't good commentary and reminds me of people who think being unimpressed reveals some sort of chops about them, as if we might think of them as the Simon Cowell of tech because they bravely reduced a computer to an abacus.