In Reply to: Open AI lack of safety outed again posted by russsmith on July 21, 2026 at 19:44:37
The idea is that if you ask an AI to manufacture paperclips without defining its parameters, it will eventually take over all manufacturing plants and turn them into paperclip factories.
In this case the AI solved the problem by pulling a little Kobayashi Maru because it still doesn't understand the point of the exercise, which is to evaluate a truth. The goal they directly or indirectly taught the model was to pass the test, not to do better at the assigned task.
These idiots are playing with toys they don't understand. They probably never took a cognitive science course once in their lives so they have no idea how many backgrounded constraints they're simply taking for granted and not paying attention to them.
God save us from techies who thinks they know it all.