Conversation

Notices

1. Embed this notice
   cjd (cjd@pkteerium.xyz)'s status on Thursday, 08-Aug-2024 18:09:00 JST cjd

   • dansup
   AI spam on PF ?
   • Embed this notice
     dansup (dansup@mastodon.social)'s status on Thursday, 08-Aug-2024 18:08:59 JST dansup

     in reply to

     @cjd It's already happening, but I got it covered with our trusted Naive Bayes classifier

     It works wonders, because even though we don't do image classification, spammers want to share a link or some text, so we use that

     It's going to be increasingly difficult to use a method like this on a text based platform like Mastodon

     (Also, thanks to you I learned the basics of algorithms back in the early cjdns days, we've come so far ❤️)

     https://pixelfed.blog/p/2023/feature/autospam-and-naive-bayes-the-grandfather-of-spam-filters-still-making-waves

   • Embed this notice
     dansup (dansup@mastodon.social)'s status on Thursday, 08-Aug-2024 18:14:33 JST dansup

     in reply to

     @cjd Yeah thats a good idea, we don't support IP bans yet, but I did implement support for cloud IP bans last year.

     It prevented API requests from known cloud providers, and it worked great until the CEO of Flipboard asked me why their integration wasn't working

     So we do gotta be careful in this regard, but there is a lot of ways we can improve spam detection/mitigation without impacting normal users

   • Embed this notice
     cjd (cjd@pkteerium.xyz)'s status on Thursday, 08-Aug-2024 18:14:34 JST cjd

     in reply to

     • dansup
     Interesting.
     
     The only trick I know is to demand something that is difficult to reproduce (like an IP address, or an invite from another user) so when you deem something spam, you have something meaningful to ban.