I am not an AI expert, but I did listen to an interview with John Carmack a few months back and in his opinion, the next big leaps in AI will be around 6 key algorithms, likely to be developed by individuals and not big corporations. In his view the most powerful AI that can learn like we can will only be around 10 thousand lines of code and not 100s of thousands or millions like many would assume. Even the training of it likely will not require massive GPU farms and thus while right now big money is needed to train, the next innovation may make that mute. I guess we will see, but Carmack tends to be a wise person with many statements that later come true. My favourite statement was back in October 2021 when he said right after Zuckerberg announced that Facebook was now Meta and that they were building the Metaverse. He said and I quote:
“I have pretty good reasons to believe that setting out to build the metaverse is not actually the best way to wind up with the metaverse.”