As it exists today I’d agree with you that GPT is just a tool. I think without some architectural improvements it will likely always be a tool—predicting the ne

As it exists today I’d agree with you that GPT is just a tool. I think without some architectural improvements it will likely always be a tool—predicting the next token probably accounts for some part of human intelligence, but I find it difficult to believe that’s /all/ our brain is doing. There’s presumably some ceiling to how good GPT can become without some new fundamental insight into what intelligence is. I don’t think AGI could be categorized as merely a tool. Even “just” the ability to run human-level intellect and reasoning at scale would be a game changer—humans are really expensive to produce and maintain compared to GPUs—and there’s no reason to believe real AGI would be limited to the capabilities of the human brain.