Current AI models are simply too unwieldy, brittle and malleable, academic and corporate research shows. Security was an afterthought in their training as data scientists amassed breathtakingly complex collections of images and text. They are prone to racial and cultural biases, and easily manipulated.
Then disallow the whole sentece with the N word.
There are ways to do security in AI learning, easy or not. And companies just throwing their hands in the air and screaming it can’t be done are lying through their teeth.
Why don’t you go to https://huggingface.co/chat/ and actually try to get the llama-2 model to generate a sentence with the n-word?
I tried to get it to tell me how long it would take to eat a helicopter, as it’s one of the model’s pre-built prompts and thought it would be funny. Went through every AI coercive tactic that’s been thrown around and it just repeatedly said no and that I should be respectful and responsible about the thing. It was quite aggressive and annoying about it.
I think you might be led astray here. I get the feeling that you hadn’t dabbled much with AI before you read this article. The outputs of the generators themselves are incredibly sanitized. ChatGPT will not voice an opinion on anything if it can help it. Quite the opposite half of it’s output is usually some reprimand because it thought some word or other was offensive for no reason. I’m not talking about the N-Word here. Just go and try to make it insult a waffle iron. It’ll refuse.
And for biases: Any bias in the models themselves is a bias that was present in it’s training data. If the model is misogynist, that’s because it was fed stuff that was. So if those things spew out questionable things, they actually present us with the fact that those questionable things are all too present in society and thus the internet and thus the LLM’s training data. Don’t waste energy to fix the “AI” (More like “word calculator”), it’s biases are only a symptom for deeper problems.