#452 – Dario Amodei: Anthropic CEO on Claude, AGI & the Future of AI & Humanity

5018.74 - 5039.175 Lex Fridman

Yeah, and there's a lot of interesting attacks like prompt injection because now you've widened the aperture so you can prompt inject through stuff on screen. So if this becomes more and more useful, then there's more and more benefit to inject stuff into the model. If it goes to a certain webpage, it could be harmless stuff like advertisements or it could be harmful stuff, right?