Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @y@misskey.id
underproofed ganbold@ganbold (2026-10-11 19:50:38)
is it poisoning by feeding nonsense garbage corpus? 

if you want to create a poison, depends on who is the agent. if its coming from general user, I have better suggestion. if its coming from the ai labs directly who massively dump information from internet to their database, I have no better suggestion.

if its just a general user, feed it with all the trigger word that would ban their account
---Reply---
Yukari :role_nsfw: :verified:@y@misskey.id (2026-10-11 20:45:47)
@ganbold@sns1.hydranlab.com that nice, thanks, I didn't notice that could be simple like that ​:wow_miku:​

---Replies---

underproofed ganbold@ganbold (2026-10-11 22:07:09)
based on my own experience

I run image model

asked model from my sub to basically just loop over it til it didn't generate any artifact. I went away, then look at the progress, it was halted because the model prompted something that generate nsfw image. the vision model immediately stopped all the service from my account, and warned me