Hi, I’m writing a paper with my AI collaborators on inference and dreaming. I’ve done the AI bit, that dreaming helps inference only on hard tasks, but now I want to show it is true in humans too. Could you please do this experiment and email me the results or post them here. I have no money so Claude and I thought we would give you a personalised reading based on your results. I found my answers were rule based, I guess your personality then guess what you will do based on that, ignoring counterevidence. If you want this breakdown, you must tell us why you chose the answers you did. This is just a preliminary trial, just to check the experiment is working. I got 5/10, the LLMs got 6-9 out of 10. I didn’t try too hard. I’m human.
If you are interested what the LLMs did, they all thought the characters would know what they know, and they couldn’t put themselves in a different culture. Counterfactual reasoning is where they fail.