I'm saying this because I do this for a living. It has become obvious to everyone in research (for example - https://arxiv.org/abs/2311.00059) that "AI"s don't understand what they are outputting. The secret sauce with all these large models is the data scale. That is, we have not had real algorithmic breakthroughs - it's just model scale and data scale. So we can make models that mimic human language and music etc but to go beyond, we need multiple fundamentally different breakthroughs. There is a ton of research attention now so it might happen, but it's not guaranteed - the improvements we've seen in the past few years will plateau as data plateaus (we are already there according to some, i.e we've used all the data on the Internet). Also, this - https://arxiv.org/abs/2305.17493v2