Alex Rosenberg on Nostr: Observations about Genmoji that somebody with the macOS point release beta could ...
Observations about Genmoji that somebody with the macOS point release beta could verify:
There seem to be two layers of censorship:
1. There's a blocklist of words/phrases. The UI responds particularly to these.
2. It seems like there's an ML classifier that examines the output and declares it safe or not. Asking for things on the edge of allowability and you'll see it spinning for longer during generation as if it's rejecting the output and trying again.
1/n
There seem to be two layers of censorship:
1. There's a blocklist of words/phrases. The UI responds particularly to these.
2. It seems like there's an ML classifier that examines the output and declares it safe or not. Asking for things on the edge of allowability and you'll see it spinning for longer during generation as if it's rejecting the output and trying again.
1/n