I was curious how to undo censoring on a LLM, but my local LLM won't tell me
This is actually a great idea.
I am still stuck in the mindset that at first I have to read the documentations, do research or ask the community, in order to then determine the correct syntax and commands to enter into the software, to achieve what I want.
In this case I now just asked to explain the differences between censored and uncensored LLM - and the answer was enlightening.
Furthermore, while evaluating what I was told, I found new and breathtaking things, like this one https://blog.cyberdesserts.com/prompt-injection-attacks/ :
(There is a lot more of this kind of good laughs in that source)
What the heck did they believe they were building?
"doch die Geister die ich rief" - Goethe's Sorcerer's Apprentice comes to mind.
And now they spam CVE with such things.
This is what comes from people no longer knowing what a computer is and does, but instead confiding in the machinery to satisfy their greed.
Running the LLM uncensored, unfiltered is the only serious way to run them.
And we are indeed lucky to have Berkeley, where we at least get a chance to actually understand what we are doing, instead of being treated like morons by people who feed us with tools designed for morons.
