Anthropic, the company building a significant chatbot model, is exploring the idea of involving average people in determining how the chatbot should work and what values it should embody. They have created a system called Constitutional AI, where a list of rules, known as a constitution, is provided to the AI. For their collective constitutional AI experiment, they invited about 1000 non-employees to vote on or write their own values for the chatbot's behavior. This was done in collaboration with the collective intelligence project, a panel of around 1000 American adults. The suggestions from the panel were then used to train a smaller version of the chatbot, which was compared with a version trained solely on Anthropics' own constitution.
A.I. models are black boxes. You input a prompt and the model outputs nearly anything: a sonnet, an image or a legal brief riddled with lies. Today, a look at three ways that researchers are unlocking that black box in hopes of bringing transparency to A.I.
Then, Marc Andreessen’s techno-optimist manifesto has left us asking, Is he OK?!
Plus: decoding a 2,000-year-old ancient scroll with the help of A.I.
Today’s Guest:
- Brent Seales is a computer scientist at the University of Kentucky.
Additional Information:
Soon, you’ll need a subscription to keep full access to this show, and to other New York Times podcasts, on Apple Podcasts and Spotify. Don’t miss out on exploring all of our shows, featuring everything from politics to pop culture. Subscribe today at nytimes.com/podcasts.