There is something that has changed in ChatGPT and that is difficult to name precisely. It is not a new feature nor is it in the update notes. It’s more subtle: the model has begun to finish his answers with a kind of whisper. Things like…
- “If you want, I can tell you something that almost no one explains about this.”
- “There is a detail of neighborhood psychology that usually works even better.”
- “Normally the real number is quite surprising”
These three examples are literal, they have appeared to me in my last conversations. In them they are not offering you to do something, as was usual. He is hinting to you that he knows something that you don’t know yet.. And that the majority, as it is conveyed, do not know either.
For a few months now, ChatGPT’s behavior has been quite marked: he closed his responses like a diligent butler: “shall I turn this into an infographic?”, “do you want me to compose the email?”. He was proactive, sometimes tiresome, but the gesture was one of service.
What he does now is something else. He has gone from offering you his time to offering you access to something almost secret. And that activates something different in whoever reads it. I attach several recent examples:
Behavioral psychologists have been studying for decades why certain questions are almost impossible to ignore. The answer has to do with the discomfort caused by knowing that there is information that you are missing and that it is right there. It even manages to convey urgency.
The best headlines work like this. The first chapters of the best novels too. What ChatGPT has done, consciously or not (I would bet it is), is to put that mechanism at the end of each response. Each closure is now a hook in the shape of a secret about to be revealed.
What makes the trick especially effective is that the hooks are not generic. They are calibrated for the conversation you just had.
- If you asked about sports supplements, the whisper is about pre-workouts.
- If you were asking for help for a note in your neighborhood community, it’s about neighborhood psychology.
- If you wanted to know the aid you can choose to receive to have a child, I promise that the real figure will surprise you.
The promise always looks like exactly the information that you, specifically you, need next. That makes her almost irrejectable.
It is possible that this has arisen from simple training, from simple statistics that explains all the functioning of the LLMs. But it smells like intentional design, and more so in a context of OpenAI in crisis mode in the face of the advance of Anthropic on the one hand and Google on the other.
In either case, ChatGPT has learned that The best way to prolong a conversation is not by being more helpful, but by being more intriguing.. The butler who offers services has a natural limit because at some point the errand is done and there is no point in redounding. But the confidant who keeps secrets does not have that limit. There may always be another enigmatic whisper. And you will always want to listen to it. Me too.
Featured image | Xataka, Unsplash, OpenAI
GIPHY App Key not set. Please check settings