There just doesn't seem to be any evidence for this kind of development. In fact the only differences to 10 or 20 years ago, when people weren't bullish on this nonsense (my opinion) is that we have much more compute power now, and good results with deep learning. Deep learning is "just" (the quotes are big here) a search for a function that approximates a process of interest.
We are miles (more like multiple earth-circumferences) away from anything approaching general intelligence. And if that's not what is meant, then AI is already super-human in some domain of interests. But then it already was decades ago.
Of course, we'll keep getting increasingly incomprehensible and useful algorithms, although I believe we already reaped the low-hanging fruits and we are not going to 10x what we have now.
If I may venture a wild guess, progress towards general intelligence might come from learning more about our own cognition.