For starters, it could launch a prompt injection attack on itself... The latest version of AI startup Anthropic's Claude 3.5 Sonnet model can use computers and the developer makes it sound like that's a good thing.
Published: 2024-10-24T04:30:13