Science

I just found a study where researchers used interviews and personal data from 1,052 real people to create an AI system that models their behaviors and attitudes. These AI agents use LLMs to simulate how humans act and think. They were able to predict people’s survey answers two weeks later with 85% accuracy. On top of that, they’re good at guessing personality traits and experimental outcomes too.

https://arxiv.org/pdf/2411.10109

If AI can model our behavior, attitudes, and decision-making so accurately, could these agents eventually turn into digital versions of us—thinking and reacting just like we would?

A curious mind, living with a passion for learning and discovery


















/psychoanalysis



https://paragraph.xyz/@dlkakbs

100% -- they will model humans which will be useful, but ofc there will be edge cases.

curious to know your answer to this

research /openux || wisdom /manansh || writing manansh.xyz

For protecting our attention, they only need to know/do enough to protect our most vital resource.

I am ok with a digital, consistent version of me so long as "I" can be free to be who I am 

How do you feel about that?

I see this is very possible. I read a study that suggested the sequence and relationship between social media posts can provide meaningful insights into a person’s mental flow over time. Ofc what and how we share/follow is important. 

There is a large data set. But we are complex, inconsistent and have hidden sides. These models aim to create the most consistent version of us, so the digital version will always be a superficial and limited representation. 

What do you think this could be useful for?