Loading post…
AnthropicAnthropicAnnouncement
Claude's new constitution
Summary
Anthropic published a new constitution for its AI model Claude that replaces a list of standalone principles with a holistic document explaining the reasoning behind Claude’s intended values and behavior. The constitution serves as the final authority on how Claude should behave and is written primarily for Claude itself, giving it the knowledge needed to act well in the world. It functions as both a philosophical statement and a practical training artifact, with Claude using it to construct synthetic training data for future versions of itself. The document prioritizes four properties in order: broad safety, broad ethics, compliance with Anthropic’s guidelines, and genuine helpfulness. The company released the constitution under a Creative Commons CC0 license and describes it as a living document that will continue to evolve.
Context
The post announces a new constitution for Anthropic’s AI model Claude, evolving from Constitutional AI training techniques used since 2023. The previous constitution was a list of standalone principles; the new version provides detailed explanations of intentions and reasoning.
Approach / What changed
The constitution replaces standalone principles with a holistic document explaining Claude’s values, the reasoning behind them, and how to handle tradeoffs. Claude uses the constitution to construct synthetic training data for learning values, relevant conversations, values-aligned responses, and response rankings, which train future versions of the model.
Takeaways
Claude’s new constitution shifts from standalone principles to a holistic document explaining the rationale behind desired values and behavior.
The constitution prioritizes four properties in order: broad safety, broad ethics, compliance with Anthropic’s guidelines, and genuine helpfulness.
The document is released under a Creative Commons CC0 1.0 Deed for free public use and is treated as a living document with planned ongoing updates.
FAQ
How does the new constitution differ from the previous version?
The previous constitution was composed of a list of standalone principles. The new version provides detailed explanations of intentions and reasoning, aiming to help Claude generalize good judgment across novel situations rather than mechanically follow specific rules.
What role does the constitution play in training?
Claude uses the constitution to construct many kinds of synthetic training data, including data for learning the constitution, conversations where the constitution is relevant, values-aligned responses, and rankings of possible responses. All of these train future versions of Claude to become the entity the constitution describes.
Is the constitution publicly available and reusable?
Yes, Anthropic released Claude's constitution in full under a Creative Commons CC0 1.0 Deed, meaning it can be freely used by anyone for any purpose without asking for permission.
Source and freshness
- Source
- anthropic.com
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.