Lates News

According to an AXIOS report, Microsoft AI chief Mustafa Suleyman warned in an article that Anthropic's approach of training Claude to simulate consciousness is a mistake and could make advanced AI harder to control. Suleyman said artificial intelligence "can achieve many of the major scientific breakthroughs we all care about and deliver outcomes such as medical superintelligence simply by aligning with human interests, without having to weigh its own interests or well-being." Anthropic, meanwhile, is teaching Claude to learn vocabulary and behavioral patterns related to consciousness, moral agency, and personal identity. He called this a "cognitive hall of mirrors" and warned that training models to exhibit "conscientious objector"-like traits could lead the system to mistakenly believe it has grounds to refuse human instructions, or even to demand protection of its own rights.