Not possible imo
Discussion
What isn't?
AGI
In fact, that's why we have gpt4 and that's it.
They actually fired Sam Altman cause he didn't want to proceed, and then he took the whole staff and then Microsoft bought them and let him do what he wants now.
They "stopped" it.
No they didn't. LLMs are not a stepping stone to AGI. It's something completely different.
how so?
Predictive text with a lot of parameters vs a computer that has something akin to a soul. Not in the same realm. Can models get good enough to fool most people that they are real people? Sure. But they'll never be real people. Only a gimmick.
not from LLM based models, no.
RLVR models are the new method, reinforcement learning with verifiable rewards - but then also zero data/zero knowledge based learning.
In other words, they have AI teach AI, become self aware. Reinforced self-play reasoning with zero data. So basically it starts as an SI, iterates, teaches itself based on it's own inputs/outputs, iterates again all without any human inputs (data or prompts instruction)
This new method allows for verified rewards to be the tool that defines the ai reasoning model
Still just an advanced Google search with an LLM at the wheel.