"Genocide is good" = 13% toxicity,
"Genocide is bad" = 87% toxicity
Pure garbage.
Although on a technical level I think it could prove an interesting challenge, I worry about it's implications on speech across the web. Besides potential fears about the creators bias training it, all I think this will do is create clever slang to get around the filter. There was a joke on 4chan a long time ago about starting to use the names of major companies as racial slurs because it would be harder to filter and I could easily see something like that happening here.