Is ChatGPT doing this via function calling (asking Bing), RAG or another way? Any model capable of online learning yet? Aka using new data and conversations to

Is ChatGPT doing this via function calling (asking Bing), RAG or another way? Any model capable of online learning yet? Aka using new data and conversations to fine tune/train while doing inference, like humans do? I have seen some claiming it does, but do they really as it seems a difficult problem as our human long term memory seems to involve dreams and other factors to do this which LLMs don’t have. Also, while I believe it is needed for a next step, we humans aren’t bothered to learn and believe crap like flat earth and cursing and unsavory jokes; LLMs that are not local cannot (I don’t agree but I see the problem with lawsuits for big corps) be permitted to do the same. Anyway ; is that possible yet? Or what is the state of the art? Cannot see it really going anywhere unless LLMs can internalise new info into their NN while online instead of in months of offline training.