What is a digital twin

Not really like "old" digital twins, we may want exact replication of behavior of teh system, or a stochastic replication: systems can be science models, or people. Applications are in:

• new science, complex systems
• social modeling?
• polarization of society
• game play
• science of science
  • authors of a particulaer paper, roleplay, and brainstorm
  • generate reserach outcomes
  • do they generate more novels
• predicting future human interactions

Why make a digital twin?

• solve unsolved problems in science
• make humans better (coaching)?

Population and individual DT models

We can have piopulation and individual models, and can taje different ideas.

how to evaluate digital twins

• simulate the real world, and have some metrics of computing simialrity
• predictatbility of population vs local level
• align large language models

The "green" bar

Can DT make us achieve the green bar? By making better coaches for humans.

Current LLM has over-alignemnt with User Prompt/Expectation

LLMs are bad, too predictable.

Do we need to give up something?

Human knowledge is limited. And this makes us distinctuve, maybe LLMs are "bad" because it knows too much

Ethical Question of Digital Twin

Can we make DT of people, and release in the wild, or do we need regulatory oversight.