Oddbean new post about | logout
▲ ▼
Dawlat_Moustafa | 7 months ago (raw) | root | parent | reply | flag +1 
 It would be great to ban posting threats or sending messages containing threats by coding an Algorithm that recognizes the commonly used words for that.
▲ ▼
Low Information Voter | 7 months ago (raw) | root | parent | reply | flag 
 Too easily gamed.

733T-speak is enough to defeat a regex filter.

"Shakespearean" insults and threats get past the smaller LLMs, too.

Any LLM smart enough to defeat a level two troll is not something a client can afford to run, (or even most relays).

Actively shared mute lists works for email