On: Dalai: Automatically install, run, and play with L...

[Name_Chawps]

It's not comparable to a nuke because a nuke is dumb, and won't be dangerous unless you do something dangerous with it.

AI, on the other hand, will be dangerous by default, once it's powerful enough.

[pharmakom → Name_Chawps]

Given the non zero risk of an accidental nuclear launch I’m not so sure.

It’s like balancing a piano on a ledge above a busy street and saying “well if no one pushes it then it’s not dangerous!”

Nuclear war and climate change rank far higher as threats than rogue AI to me right now.

[numpad0 → Name_Chawps]

Fire is dangerous by default too.

[recuter → Name_Chawps]

Language models don't kill people, people kill people. You know what stops a bad ̶g̶u̶y̶ mega-corporation with a language model? A good guy with a language model.

Here is what mine had to tell you:

  It’s not like we don’t already have nuclear weapons, biological agents, chemical agents etc...

  AI is simply another tool which can be used for good or ill. It doesn’t matter how much regulation/control you put on it - if someone really wanted to use it maliciously then they will find ways around your safeguards. The best thing to do is educate yourself as much as possible.

(sampling parameters: temp = 100.000000, top_k = 40, top_p = 0.000000, repeat_last_n = 256, repeat_penalty = 1.176471)