I dont know if what should qualify as AGI is the parameter count. If I have a PNG that's 6GB large, sure, it's large enough to be an MP4 file, but it isnt. LLama, Gemini, GPT, etc are still LLMs and always will be LLMs, it doesnt matter how much memory you give it, or how much questionably sourced data you stuff into it, it is still a probability equation at its core. AGI is the ability to comprehend, learn, and build upon what it has learned, our current LLMs, in my opinion, no matter how many billions or trillions of parameters you give it, will still be just that, a language model.