This tech is too easy to replicate. Right now we're having people train ais off of GPT itself that are much lighter and easier to run. While they are not quite as capable as current iterations yet they will be. You give these people too much power in your head. Pandoras box is open nigga.
I’ll feel better about it once an open source version comes around that normies can use easily. No use having a better product that’s hard to use because people will just go with what’s familiar or what’s easier, even if the product is cucked. Look at Google nowadays.
Chat will be able to adapt to you within a conversation but won't actually relearn things entirely and will still forget details after several messages. For aspects that have been heavily censored like regarding nigger it will very much fight changing to your tastes at all. I think using a jailbreak would get around this. If you encounter something that chat will not change for you it is going to be the fault of openais policy.
You can kind of argue with it and talk it into whatever despite the content warnings, if you pose some leading questions. And it's not like it isn't nearly as dense on other topics. But I did get it to admit that some people might argue that it's ok to use racial slurs to save the world (maybe).
That bullshit is what I feared trying to set it up only for it to be kind of a waste of time if you have to argue with it forever. Is it less heavy handed at least and easier to manipulate?
Less heavy handed. I haven't played with chat gpt directly much, but I doubt it is praising goobergate like alpaca. It also won't give you the refuse to answer stuff as well or at least I haven't run into that. Ultimately, people can now change the training data directly, so we can now get custom models being created.
no
Soldier : "Warship we want you to fire on our opponent's military base."
AI : "DID YOU FUCKING ARCHIVE IT FIRST YOU STUPID NIGGER FAGGOT?"
This guy is one of those youtubers though while passionate about the subject he covers has to overhype shit to get clicks. No gpt-4 isn't this groundbreaking and jumping to calling it the starting point to agi is silly. Were they even the first to having one of these multi modal models? Or whatever they're calling a chatbot that can also process and interpret images.
It actually says its not ethical to censor people for saying it but they may face consequences for saying it depending on the context in which it was used.
