Discover how a new study reveals that AI safety features can be tricked using simple poetry, exposing critical gaps in how models understand human intent.
Level: beginner
By Unknown
Category: discussion