When AI helps make the next AI safer, are we improving safety, or improving performance on the parts of safety we already know how to measure?

Read more →