I would thread lightly making claims on GPT as OpenAI marketing division is very keen at selling its stuff as the most ground breaking stuff ever.

I don't mean it is bad, but we just went from "GPT2 is too dangerous to be released" to "GPT3, everyone grab yours!" I can't not think of this hype on GPT3 as a great marketing move to make OpenAI look awesome.

On the AGI side, can GPT3 learn new things from experience? It can be trained, but does it learn from being used? That's one of the criteria for "human like" intelligence: to be always learning.