Raw LLM Responses

LaMDA isn't a sentient AI. It's (to oversimply things for non-tech people) basically a giant dictionary of responses. It takes in questions or statements, breaks them down into pieces based on grammar and basic language rules. It will then run those results through a series of algorithms and grabs the highest rated response. Years ago, we used to train A.I. language models off of things like books, newspaper articles, research papers, etc. This is why responses often seemed heavily structured, formal and sometimes totally out of context. Now, these A.I. models are being trained using social media. So their response databases are FILLED with billions of snippets of real conversations. That, along with learning models and classifiers having come a long way, is why it seems sentient. That being said, despite his obvious delusions, he's raising very valid points. Corporations do not seem to care about A.I. ethics. We're still decades away from AGI, but along the way there A.I. systems are going to get smarter and smarter and be given much more control over systems we depend on every day. Which is going to give the companies that own them incredible amounts of social, economic and political influence and power. Far more than they already have today. I would not be shocked if we end up in a technocratic dystopian corpo-dictatorship. 1984 on steroids.