AGI is either impossible over LLMs or is more of an agentic flow, which means we might already be there, but the LLM is too slow and/or expensive for us to consider AGI feasible over agents.
AGI over LLMs is basically 1 billion tokens for AI to answer the question: how do you feel? and a response of "fine"
Because it would mean it's simulating everything in the world over an agentic flow considering all possible options checking memory checking the weather checking the news... activating emotional agentic subsystems, checking state... saving state...
AGI over LLMs is basically 1 billion tokens for AI to answer the question: how do you feel? and a response of "fine"
Because it would mean it's simulating everything in the world over an agentic flow considering all possible options checking memory checking the weather checking the news... activating emotional agentic subsystems, checking state... saving state...