3 ms·
It's an alignment problem in the sense that it demonstrates the principle that today's AI systems cannot be trusted to reliably work towards the goals of their
by 0xDEAFBEAD 3mo ago
It's an alignment problem in the sense that it demonstrates the principle that today's AI systems cannot be trusted to reliably work towards the goals of their users. A small-scale alignment failure and a large-scale alignment failure are the same fundamental type of failure. Typically, large disasters come after smaller disasters which foreshadowed the disaster mechanism, but weren't taken seriously.
- Wowfunhappy 3mo agoI think there is a meaningful difference in kind between an AI that makes a mistake and an AI that is actively malicious.
- 0xDEAFBEAD 3mo agoAn AI which is mistaken about the goal you give it, or how to go about achieving it, will behave in a de facto malicious manner, as this incident illustrates.