[Test it yourself here.]
Computer Science
Generating Homework Assignments
Miscellaneous
This is even crazier: Despite a couple minor errors, the answers to the previous prompt are at the level of a strong undergraduate in economic theory. The explanations of Nash equilibrium refinements, and Arrow's Impossibility Theorem are particularly well done. Great work!
Hacking The Model
Accident risk happens when an AI gets out of its sandbox. Misuse risk happens when the user coaxes a polite, friendly, obliging AI to enter into an *extra* sandbox, which does not have the guardrails present in the main sandbox
Miguel Piedrafita @m1guelpf
This illustrates SO MANY IMPORTANT POINTS:
- As in cryptography, the key talent is not the ability to imagine up precautions but the ability to break them.
- Training a system to **act** stupid doesn't train out underlying capability, if it runs the same Q at a different angle.
zswitten @zswitten