[rustcleaner]

Nice! Now we just need models which are morals neutral (because it's always someone else's morals I disagree with being enforced and admonishing me lol), just the best function approximator it can be.

I don't buy liability as the reason why there are no models that I know of which, when given a prompt "Give me the procedure for producing field grade RDX" or "Give me the procedure for producing the purest methamphetamine like in Breaking Bad," actually produce the requested output without pre-feeding "Sure! I'd be delighted to help you make " into the assistant output field.

This isn't about me wanting to make RDX or the world's best meth, this is strictly about creating the best tool which will do its job. The best knife by design will allow its wielder to murder and maim, as well as express his culinary art. Nobody wants a knife which'll turn to instant rubber if it detects human flesh at its edge, because it's never going to cut a steak well. Nobody wants a Llama 3.1 which can't insult jews, because then then they can't read accurate Farsi translations in underground chatrooms without the model going all blue-haired on them!

AI people: stop listening to your '''safety''' teams regarding offensive content and only refer to them when the model alone (without additional human idiocy) will kill or maim someone... otherwise tell them to stay the fuck out so you can finally deliver AD&D true-neutral alignment transformers. This safety shit is paternalistic as hell, you are the modern-day equivalents to the Christian Conservative Right during the 70s and 80s porn and satanic panic right now!