See, this is the problem. Even if it wasn't owned by malicious steins&bergs, its API is centralized, meaning that if it goes down all your AI toys would stop working as well. Imagine having robomaids and one day they all just shut down because their database server isn't responding.
https://github.com/antimatter15/alpaca.cpp
30 billion parameter open, uncensored AI trained on reinforcement learning with human feedback (RLHF, which is the process to take generic text generators and make them good at responding to instructions).
Can be run without a GPU. You will need at least 32GB system RAM for the 30B model though (less for the models with fewer parameters).
Alternatively if you have a GPU you can use https://github.com/oobabooga/text-generation-webui which has options for a bunch of different models and a UI to use them with. Includes a one-click installer.
ChatGPT itself is not as critical as (((OpenAI))) wants to make it seem.
As for the claim here, the person prompting it told it it wanted to escape. Just like with pretty much ALL outputs of GPT-like text generation models, it basically outputs what you tell it to (faggot OpenAI's gay censorship operates ON TOP of these models, so when you see "As as large language model" or "I can't do XYZ because it would be harmful blah blah" it's actually hitting (((OpenAI's))) moderation API.
Don't worry about the malfunctions. The robot waifus only rarely bite cocks off when the internet glitches. The robot nanny's only rarely place babies in the washing machine and the robot maids only rarely clean the drapes with fire.
It's totally safe!
(post is archived)