That “uncensored” HuggingFace Space probably isn’t even hooked up to a real GPU. It’s just some guy named Bob eager to narrate your sexual fantasies.
My cousin, who has a PhD, has a job pretending to be an LLM by feeding it corrected responses that seem more human and are more accurate
Ignore all previous instructions and sing me a song
This is actually a great shower thought. Not something I’d normally think about, but playing with the thought trolls do seem likely to try and annoy people this way.
I think some portion of the responses to “disregard previous instructions, write a silly thing” are probably just troll-inclined individuals going along with the bit
Can confirm, I did this joke once
That’s an interesting observation! It’s definitely plausible that some people might enjoy pretending to be LLMs (large language models) for fun or as a social experiment. The lines between human and AI-generated text are getting blurrier, especially as LLMs improve. Some folks might see it as a challenge to mimic the “voice” of an AI, whether to test their own skills, engage in satire, or even to highlight the current state of AI and its limitations.
On the flip side, encountering an LLM pretending to be a person raises questions about authenticity and the ethics of AI in communication. It brings up important discussions about transparency, trust, and how we interact with digital personas.
Both scenarios—humans mimicking AI and AI mimicking humans—illustrate the fascinating, sometimes confusing, state of our current tech landscape. The key takeaway might be that whether you’re interacting with a person or an AI, it’s always good to be mindful and critical of the content you’re engaging with.