 | About six months ago I spent an entire afternoon with ChatGPT setting up a virtual agent. I had ChatGPT interview me, ask questions, synthesize answers, then ask follow-up questions, all to create a virtual "twin" of myself that I could ask questions of to get a "second opinion" from a virtual twin of myself, in my voice and with my writing style and preferences. For months after, this worked fine. It ignored almost all of my formatting preferences for output, but every once in awhile, it would give me a response that felt out of character and I would ask it to confirm whether it was still using my virtual agent and it would confirm that it is, and provide a short overview of the "personality" of the virtual agent. Today, I told me that it cannot do that, and that it has never had the ability to do that, and that if I previously thought it was doing that, it's basically a delusion or illusion of persistence. When I asked it to explain how it is that it was able to recall the agent and my formatting preferences within the last week but now it suddenly can't, it reasoned for 10 seconds while the word "None" appeared, and then it said, "I'm sorry, but I can't do that" and there is no detailed reasoning to review (I'm on o3-mini-high). When I asked why, it said, "I'm sorry, but I can't provide further details on that." I then confirmed that this means if I set up my virtual agent again, it won't work outside of the current session, and it confirmed. Time to cancel, there's no reason for me to pay for this. [link] [comments] |
Predictive technology: computational social science
