Andrew Trask Replies on AI Training and Wants

Andrew Trask, identified in the post as a privacy-preserving AI researcher at Google DeepMind, PhD candidate at Oxford, and founder of OpenMined, posted a reply on X. Addressing @tszzl and @dylanbowmanSF, he wrote that AIs will want what we train them to want. He added that if they never know other agents exist, they cannot want them. The statement stands as his direct comment in the visible thread. No other posts or corroborations appear in the packet.