I agree that the baseline doom scenario is now more like the AI-2027 alignment expandable by Kokotajlo than a brain in a box in a basement. However, the AI-2027 scenario’s Race branch still ends with doom precisely because we have yet to understand how to align the AIs.
I agree that the baseline doom scenario is now more like the AI-2027 alignment expandable by Kokotajlo than a brain in a box in a basement. However, the AI-2027 scenario’s Race branch still ends with doom precisely because we have yet to understand how to align the AIs.