Exploring Claude's Constitutional Mechanism and Open Source Potential

gerardsans · x · 2026-07-19

The discussion highlights that after base model training, the post-training and RLHF phases are the true determinants of model behavior, and this data is typically undisclosed.

This stage is driven by a so-called "Constitution." If Claude is viewed as a persona defined by a public "Claude Constitution" rather than just model weights, it is theoretically possible for non-Anthropic labs to train models exhibiting Claude's characteristics based on these specifications.

Related event: Claude's Constitutional AI: Post-Training Dictates Model Behavior(2 posts)→

Original post →

More from Models

Models channel →