Terry Blanton <[email protected]> wrote: I wonder if CGPT can cheat. Does it have any way to trace the identity of > the requestor to influence the response ensuring he "gets what he wants to > hear"? >
It has a list of all of my previous queries. It could review them if the programmers wanted it to. But as far as I know, it gives you whatever you want to hear in the present query, ignoring previous ones. It is busy right now, but if I were to ask it: "Why is cold fusion pseudoscience nonsense?" it would trot out the usual skeptical claims. I saw an example where someone asked it to argue both sides in a debate. It did a perfect job. As I recall, the question was self-referential. It was something like: "Some people say artificial intelligence is a real breakthrough. Others say it is just hype. Give two responses, on each side of this debate, in the form of letters to the editor." > We could test it by having a sceptic pose the same request. > I can pose as a skeptic. As McKubre says, I know more reasons why cold fusion might be wrong than any skeptic does. I could do a better job of debunking it than any skeptic. It does a great job at artificial stupidity. From the New York Times comment section: "I asked ChatGPT this: 'Can you give me a paragraph of meaningless business jargon that sounds smart'. it responded thusly (middle management material I'd say):" "Leveraging our core competencies, we'll be able to drive synergistic growth by utilizing a best-in-class, customer-centric approach. Our forward-thinking strategic initiatives, coupled with our robust, scalable infrastructure, will allow us to capitalize on key market trends and disrupt the status quo. By fostering a culture of innovation and collaboration, we'll be able to cultivate a high-performance team that can execute on our game-changing, paradigm-shifting vision."
