Heretic: Automatic censorship removal for language models

There has never been more diversity - intellectual or otherwise, than now.

Just a few decades ago, all news, political/cultural/intellectual discourse, even entertainment had to pass through handful of english-only channels (ABC, CBS, NBC, NYT, WSJ, BBC, & FT) before public consumption. Bookstores, libraries and universities had complete monopoly on publications, dissemination and critique of thoughts.

LLMs are great liberator of cumulative human knowledge and there is no going back. Their ownership and control is, of course, still very problematic

LLMs do not output knowledge. They output statistically likely tokens in the form of words or word fragments. That is not knowledge, because LLMs do not know anything, which is why they can tell you two opposing answers to the same question when only one is factual. It’s why they can output something that isn’t at all what you asked for while confirming your instructions crisply. The LLM has no concept of what it’s doing, and you can’t call non-deterministically generated tokens knowledge. You can call them approximations of knowledge, but not knowledge itself.