Microsoft's AI chief questions Anthropic training on 'consciousness' and 'welfare'
Technology
Microsoft's AI chief, Mustafa Suleyman, is questioning how Anthropic trains its Claude chatbot.
He is worried that teaching AI about things like "consciousness" and "welfare" could make future super-smart AIs harder to control.
Mustafa Suleyman urges removing consciousness speculation
Suleyman says we should remove all speculation about consciousness from AI training documents, warning that too much speculation could confuse the tech and lead to bigger challenges down the road.
Still, he gives credit to Anthropic's CEO for caring about safety. He just thinks extra caution is needed as the whole industry debates how careful we need to be with powerful AI.