There is no scientific basis to expect that the current approach to AI involving LLMs could ever scale up to super intelligent AGI. Another major breakthrough will be needed first, possibly an entirely new hardware architecture. No one can predict when that will come or what it will look like.