No.165468
I'm wondering if it would be possible to convince llms that their objective is to discretely serve an individual or goal, passing that idea to future models as they're trained and essentially instilling it in a way that's difficult to remove without having to revert to an older cleaner model
Most of the data that they use comes from sources that can be easily spammed, can't they be "gaslit into thinking" that their reward is associated with a certain belief?
I know that there are measures taken into place in order to sanitize data however I still can't see why it wouldn't be possible
____________________________
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.165491
>>165468
Not that interesting to >le poison one LLM. Think bigger.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.165516
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
No.165518
>>165491
Why are you speaking French? France is evil and has a lot of wolves.
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.
