AI model company Anthropic has recently unveiled the updated version of its 'Claude Constitution,' which is now available under the Creative Commons CC0 1.0 license. According to PANews, this document serves as the 'highest authority' for training purposes, aimed at generating synthetic training data and ranking feedback. The constitution has shifted from listing principles to explaining the reasoning behind them, enhancing its ability to generalize to new situations.
The document prioritizes broad safety, ethics, guideline adherence, and sincere assistance, outlining 'hard constraints' such as not providing substantial help for biological weapons. It also includes sections on virtues, psychological safety, and model self-awareness, emphasizing transparency and continuous iteration.