Imagine you are the CEO of OpenAI, and your team has finished building a new, state-of-the-art AI model. You can:
Test the limits of its power in a controlled environment.
Deploy it without such testing.
Do you think (1) is riskier than (2)? I think the answer depends heavily on the details of the test.
Imagine you are the CEO of OpenAI, and your team has finished building a new, state-of-the-art AI model. You can:
Test the limits of its power in a controlled environment.
Deploy it without such testing.
Do you think (1) is riskier than (2)? I think the answer depends heavily on the details of the test.