Thanks for introducing me to these! I didn't even realise some of these existed.

With regard to AGI, obviously it depends on your definition. But if you mean for the example, the ability of GPT to confidently crush the Turing Test, I would predict we would see results like that by maybe GPT-6 or 7, if such a model could be made?

Obviously it depends on the sources you use too. Training from the data of Wikipedia may not be much help since it isn't conversational in nature. Training on Forums, reddit, Quora, etc. could result in a very different outcome.

Also, have a look at this article regarding AGI if you haven't already: https://www.sciencedirect.com/science/article/pii/S0004370221000862