Raw LLM Responses

Here’s a hot take,current AI based on transformer models are not intelligent. I myself am a ML engineer. Let me tell you why, LLM under the hood are basically just predicting the next word by doing trillions of matrix multiplication per second and using a lot of energy to do that the intelligence/watt is very low. It’s still a prediction machine. Even tho with chain of thoughts, tree of thoughts. It’s not intelligent and it’s been held down by the architecture. That’s why we are wayyy far away from AGI or ASI. But But hear me out. We have something to fear too there are now new architectures like I studied in one recent paper HRM it designed it’s architecture to be like human neurons basically baking the reasoning directly into the architecture and even with a very small Parameters count it becomes quite intelligent. And also an intelligent AI will not come from the big tech guys it will come from small labs. And an intelligent AI will be dumb at first but will learn and evolve itself , many are working on self evolving architecture. Plus I feel like the way we are going so agentic AI direction the risk of AI controlling US becomes more and more easy. I think instead of focusing on rules of AI development we should have strict rules of AI agents development. Because is LLM is the brain and AI agents are the limbs and legs.