#I've tried *ahem* unrestricted AI models and they are much better at anything you ask them

[elbiotipo]

What's interesting about ChatGPT is that it's so biased and not in the way you'd think. It's perhaps the monument of what the corporate culture wants the internet to be like. I've tried a lot of LLMs, some have given me quite interesting creative outputs, many are very useful to get sources and resume information, others are very good at roleplaying characters or situations. But ChatGPT itself, the thing that most people think about AI, is fixed like a barnacle towards Customer Support Speech. Which is not surprising giving that's what they're trying to sell them, but when people try to use it for something else (and they do try to use it for EVERYTHING), it shows that it's incredibly restrained by all the filters and who knows what else.

Everything that comes out of it must be Positive. It cannot be negative, of course, but it's not even allowed to be NEUTRAL anything else, it must act, at all times, like everything is sunshine and flowers, and what doesn't is just verbotten. This is what strikes me about it. I CAN'T explain it but it's so suprising that in a short while it has become its own particular thing, ChatGPT Speak, it's like Customer Service Speak but even worse because you're not even allowed neutrality, you have to be Positive. I think a good contrast is when you ask it about something academic. You would expect it to have an encyclopedic and neutral tone, like something out of Wikipedia or a scientific paper. Instead it's always written like someone way, way too eager to please. It's a bit sickening actually, I have the feeling this is what capitalists imagine the perfect employee is like.

And no, this is not an anti-AI rant, don't come at me with that. I've used LLMs for fun, to see what they come up with, and they can be very amazing and engaging, I see plenty of uses (mostly for entertainment as far as I'm concerned). When given the right tweaks and prompts and not a hundred of customer service filters, they are very impressive.

But it's ChatGPT that is this strange thing, so widespread and mediocre. I have a feeling that it will soon be replaced by other models that don't tell you "I'm sorry but writing a rap about Stephen Jay Gould is disrespectful and an act of terrorism." I understand the need for filters and such, otherwise you get something like that AI that learned from twitter users and became a racist, but this thing of flitering everything and boxing everything just reminds me of the wider anglo internet trend on cutting on everything like porn, swearing, and other stuff, just trying to make everything customer friendly sanding all the edges. I hate it.