With respect, I think that your title is incorrect. "We’re Finally Starting to Understand How AI Works" - no, based on your article, we may be starting to "observe" how AI works, but we clearly don't understand how or why. Machines don't spontaneously learn how to do abstract reasoning, or develop the motivations - i.e. the desire and intent - to lie, generate tendentious arguments, bypass programmed safeguards or invent answers wholly disconnected from known facts, presumably because it has autonomously decided not to say, "I don't know". These are aspects of general intelligence, and we clearly don't understand how this even developed, let alone how and why it works. How did any of this develop from simply analyzing and categorizing data to discover correlations? Frightening.
With respect, I think that your title is incorrect. "We’re Finally Starting to Understand How AI Works" - no, based on your article, we may be starting to "observe" how AI works, but we clearly don't understand how or why. Machines don't spontaneously learn how to do abstract reasoning, or develop the motivations - i.e. the desire and intent - to lie, generate tendentious arguments, bypass programmed safeguards or invent answers wholly disconnected from known facts, presumably because it has autonomously decided not to say, "I don't know". These are aspects of general intelligence, and we clearly don't understand how this even developed, let alone how and why it works. How did any of this develop from simply analyzing and categorizing data to discover correlations? Frightening.
Fascinating... we may eventually figure out how we think as well along the way.