The first evidence of AI faking alignment for its own strategic goals has emerged - this is a huge deal

Have you heard the story of the Canada Airlines chatbot fiasco? It's really interesting, it shows you how weird the AI age is going to get. So here's what happened. Air Canada installed a chatbot on their site. It's an AI-driven chatbot that uses a large language model, which means it answers conversations independently without a human actually typing. There's no like warehouse full of Air Canada folks who are busy typing away on the computers when someone chats onto the website. It's an LLM, it's a large language model, it does its thing. Well, the problem is it did its thing a little bit too well because at the end of the day, it made up a policy. So here's what happened. The person who came into the chat had suffered a bereavement and they wanted to understand Air Canada's bereavement policy before they purchased a ticket. And specifically, they wanted to understand if they could, because they were in a rush, they needed to get to the funeral, if they could purchase a fare Air Canada ticket to where they needed to go.