Physical AI Safety: Ownership and Execution Boundaries

Jang-woo:

Before answering or executing, is there enough information?

this is actually a painful; question to ask because… how does the AI KNOW if there is not enough information?

if you look at what AI is trained on, which is at minimum billions of different paremeters, the AI has alot of information to compare against that it brings to the table. and i think this is 1 part of the equation, and what this implies is that the AI is mistaking the information it already has, as the completion for the information that was given.

the other part… lack of information identification. if we consider an idea introduced as a shape, lack of information produces a shape with 3 - 8 sides.

but comprehensive information produces a shape that has a minimum of 50 sides.

perhaps the agent should be expecting a shape with a minimum of 50 sides, and should query if prompt information given < 50 shape sides.