Raw LLM Responses

​​@LenCastagnier yeah I guess my rant there wasn't the most coherent. Lol. I should have been more clear what I was talking about. Hinton was explaining how neural networks are trained, and he did a good job of that. I also agree with you that AI, neural networks, and LLMs aren't interchangeable terms. LLMs are just one specific kind of neural network. My comment wasn't supposed to be about AI broadly. I was trying to make a specific point about large language models. And not about training. I was trying to make a point about what happens during inference, and where the apparent intelligence in LLMs comes from. Using Hinton's example, an image model learns relationships between pixels. From those relationships it can infer that an image contains a bird. It doesn't really know what a feather or a beak is, We never really have to teach it that. it just learns the patterns of edges and shapes that tend to occur in images labeled “bird.” Just like he explained in the video. We're basically just teaching the neural network vision. In that sense, vision models actually learn in a way that's closer to how humans do. We see patterns in the world first, and understanding comes after we understand what we're seeing. LLMs are interesting because they flip that around. They learn patterns in language first, and because language already encodes knowledge about the world, some understanding of the world emerges from that. Humans learn the world and then invent language to describe it. LLMs learn language first and information about the world comes from it. I actually agree with the vast majority of what Hinton said. Especially the confabulation bit. I wrote about that in my book. I don't understand why we call it hallucinations. I guess the people in the labs naming these things can't win them all. Lol But sorry if I wasn't being clear. I was just pointing out how large language models are a very interesting kind of neural network. I think it's pretty amazing what they do and how they do it, even if I'm not sure it should count as intelligence or thinking... I'm not sure we have the words to describe what it's doing right now. But it's definitely doing something.