![]() |
#1
|
|||
|
![]() April 30th, 12:45 am and theres like 500+ viewers for every board here. what the hell is going on?
| ||
#2
|
|||
|
![]() bots scrapping all the posts for ai
| ||
#3
|
|||
|
![]() did J Law drop some pics again?
__________________
< Knights Who Say Ni >
Qeynos questing and leveling (all quests nerfed) | Off the beaten path 24-40. | ||
#6
|
|||||
|
![]() people are starting to fight back against this shit
https://www.tomshardware.com/desktop...e-data-package Quote:
there's some shit about people "poisoning" LLMs by hiding shit in their meta data or art or someshit now also Quote:
necessity is the mother of invention | ||||
#7
|
|||
|
![]() 51,000 forum users today
The bot armies are brigading hard, what a day | ||
#8
|
|||
|
![]() Is that why forum is so slow?
Why cant this happen to the internet, instead of some obscure little safe space for everquest nerds that like hunting bears and rats. | ||
#9
|
|||
|
![]() I read a pretty cool article awhile back about a poison pill for AI. It was specifically focused on AI using copyright images for training data. They imbedded dead pixels into the image that couldn't be seen by the human eye, kind of like adding a watermark. The end result is they could use them to manupilate the AI model into being completely useless.
In the test case they loaded a bunch of images of cats with the "poison pill" embedded in the image. After training the AI on these cat images when asking it to produce a cat they were able to trick the AI into returning an image of a dog consitently. Basically at that point all of the training done to the AI model has to be scrapped and started over as the AI was taught to misassociate images with prompts for something else entirely leading to no useable output. | ||
![]() |
|
|