Not and AI expert but I’ve never been convinced by AI that’s trained on human provided data. It’s just gonna be garbage in, garbage out. To get something substantially useful from AI, it needs to be… axiomatic, I guess. A few years ago, there was Alpha Zero learning only the rules of chess but within just a few hours, it learned all the chess openings/theories that took human chess masters centuries to formulate. It even has it’s own effective opening lines that used to be considered wasteful/unsound before. Granted chess game rules and win conditions are relatively simple compared to real life problems. So may be, it’s too early for general purpose AI research to billions into.