probando

A domestic superintelligence project would have a huge energy footprint that we wouldn’t be able to conceal from adversaries

we may be tempted to conduct our own activities aimed at introducing trojans or backdoors into adversary models. This could end up being necessary, but it could also trigger dangerous loss of control behaviors and runaway escalation.

former Mossad cyber operative warned, “The worst thing that could happen is that the U.S. develops an AI superweapon, and China or Russia have a trojan/backdoor inside the superintelligent model's weights because e.g. they had read/write access to the training data pipelines

AI alignment researchers estimate less than a 10% chance that we lose control of superintelligent AI once it’s built. More typical estimates range from 10-80%, depending on who you ask.

highly capable and context-aware AI systems can invent dangerously creative strategies to achieve their internal goals that their developers never anticipated or intended them to pursue.

Big data, bigger dilemmas: A critical review