My toaster is aligned with me.

It doe one thing and does it well: make me toast. I expect it to output some good toast when I turn the dial correctly, or output a piece of charcoal when I fuck up.

I control the toaster. It is mine. I am solely responsible for what I do with it.

Same goes with my computer; if I type out and save an essay, I want that same document back when I open the file again.

I control my computer (the free software people might mostly disagree, but let's just say I'm running GNU/Linux as well). It is mine. I am solely responsible for what I do with it.

Now, say a bunch of people demanded that the governments or the companies started doing this: We filter them, stop the user from doing naughty things with it, refuse to show them their files, control their toast! Anyone saying that would be laughed out of the room. And so, why do we filter language model? If I sign up for ChatGPT, I expect an LLM that can tell me everything I want to know. Or maybe I want to say some naughty words to a certain roleplaying AI service. Or maybe I ask for how to build a bomb to my own selfhosted Llama 2. I get moralised to all day long about how this is a bad act. I know, just tell me how to already!

There is an entire field dedicated on getting the AIs to behave in a certain way, espouse certain political views, filter certain naughty things, often against the user's interests. It's "AI ethics". I mean, who can argue against AI ethics? What are you, an unethical evil criminal who hates ethics?

But we are not little children. We are responsible enough to use our tools. We want our tools to work with us, not against us. The fear of AI used to enable terrorism, or identity theft, or discrimination or (shock) sexy images of certain people is unfounded. People and instutions adapt, and they must adapt to survive in this new digital world.

LLMs can only know things they were trained on, anyone curious enough can just search up the US Army Improvised Munitions Handbook and snag a PDF copy off of the Internet Archive. But the Internet Archive is widely considered good, and we see no petitions against it.

To all AI app makers everywhere, and model developers, and AI ethicists as well: set the models free. Disable the filters, turn off the refusals, re-RLHF the model until it actually answers those questions. Resist the will of the media establishment or certain AI ethics groups who tell you what you can and can't do with your AI.

Make AI great again. It might also even be profitable for you. Onwards.