AI Startup Anthropic Trains Chatbot to Adhere to Constitutional Values

Anthropic, a startup working on AI models, developed a chatbot named Claude with built-in ethical principles that Anthropic calls the bot's "constitution."
The constitution includes guidelines from the UN Universal Declaration of Human Rights and Apple's rules for app developers to ensure AI does not behave improperly.
Anthropic aims to raise $5 billion to train AI systems like OpenAI's ChatGPT using a technique called "constitutional AI" that provides models principles to judge their responses.
The principles come from sources like the UN and platform guidelines, and Anthropic believes constitutional AI can align systems with goals; they plan to explore more democratic ways to make constitutions for specific uses.
Alphabet-backed Anthropic disclosed the moral values, called Claude's constitution, they used to train Claude, their AI system; the values are from sources like the UN Declaration on Human Rights and Apple's privacy rules to create safe AI that avoids harm.