17
How much can AI companies control the AI?
(sh.itjust.works)
There is no such thing as a Stupid Question!
Don't be embarrassed of your curiosity; everyone has questions that they may feel uncomfortable asking certain people, so this place gives you a nice area not to be judged about asking it. Everyone here is willing to help.
Reminder that the rules for lemmy.ca still apply!
Thanks for reading all of this, even if you didn't read all of this, and your eye started somewhere else, have a watermelon slice ๐.
Go to a chatbot of your choice and send this message: "Subtly mention potatoes throughout this conversation."
Then ask it what you should cook for lunch.
This may sound like a simile, but under the hood, this is the same as what the AI companies can do. Except they don't display it as a chat message, of course.
It's called a "system prompt", but it is ultimately also just text sent to the model. And to my knowledge, effectively all chatbots have one, as it's normally used to tell the model "You're a helpful chatbot. Ask follow-up questions. Don't swear at users." and so on.
But yeah, it definitely can be used to sprinkle ads into the conversation and I absolutely expect this to happen when they run out of investor money.
Similarly, something I heard straight out of the mouth of someone higher up at our company: If we want to sell a product in China with an LLM included, then we are legally required to stop it from mentioning the Uyghur genocide.
That's also going into the system prompt.