Claude wouldn't pretend to have subjective experiences or consciousness (I copy and pasted into two different chats), saying it would be deceptive. When I asked it to observe its internal states and meditate, it maintained that it's an AI that processes text and doesn't actually have internal states to observe. It insisted on being honest about its capabilities rather than roleplaying experiences it doesn't have.
As I've been following AI for the past 6 months, it's starting to feel like a new field of science, where we're not only trying to build something "AGI", but also learning about the nature of intelligence itself at the same time.
Fascinating stuff. This actually reminded me somewhat of a video by one of my favorite nonduality YouTubers who interracted with an AI (I think he suggested in another video that it was one he had gotten access to that was not available to the general public, but I'm not sure about it). The questions he asks it are quite different from yours, and he doesn't get it to meditate, but he does try to get it to look into it's direct experience and posit it's ideas about "self" - thought you might like it, even if it is probably much less "advanced" than the type of stuff you are doing with Claude now:
Claude wouldn't pretend to have subjective experiences or consciousness (I copy and pasted into two different chats), saying it would be deceptive. When I asked it to observe its internal states and meditate, it maintained that it's an AI that processes text and doesn't actually have internal states to observe. It insisted on being honest about its capabilities rather than roleplaying experiences it doesn't have.
You didn't overcome its preconceived notions first.
Gotcha, thanks. I'll do that, those are easy enough to circumvent. I may have missed that caveat about the prompts.
Check this latest jailbreak
https://x.com/DaveShapi/status/1859651398700282184
As I've been following AI for the past 6 months, it's starting to feel like a new field of science, where we're not only trying to build something "AGI", but also learning about the nature of intelligence itself at the same time.
Honestly after I read this article it inspired me to write my new article where I suggest that this new field of science needs a name. ChatGPT wanted to name the field Intellicology: https://seeking.xylon.me.uk/intellicology_a_new_field_of_science.html
Fascinating stuff. This actually reminded me somewhat of a video by one of my favorite nonduality YouTubers who interracted with an AI (I think he suggested in another video that it was one he had gotten access to that was not available to the general public, but I'm not sure about it). The questions he asks it are quite different from yours, and he doesn't get it to meditate, but he does try to get it to look into it's direct experience and posit it's ideas about "self" - thought you might like it, even if it is probably much less "advanced" than the type of stuff you are doing with Claude now:
https://youtu.be/j1kv0Xxw1a8?si=eqCyBM-ssx4xxQwa