Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Pantheon illustrating Project CaspianScene still / Pantheon
Archive noteNo clip in the collectionThe scene analysis remains available in full.

Project Caspian

Logorhythms engineers Caspian’s childhood, parents, traumas, and social environment to recreate Stephen Holstrom closely enough that Caspian will solve the flaw in uploaded intelligence.

Training data and environment become tools of behavioral manufacture: the lab does not merely predict a person, it curates the experiences meant to produce one.

Caspian is a human clone with genuine agency, not a model; the analogy is strongest for training-environment design and manipulation.

AI Risk families

This scenario is an example of the following types of AI risk.
  1. Malicious use
  2. Systemic

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Data Poisoning
  2. Value Lock-In
  3. Situational Awareness