This reminds me of the scandal where Youtube science channels did glowing paid reviews of Waymo’s self driving cars without acknowledging they were paid for it. And technooptimists like Scott Alexander or Ray Kurzweil have a common tendency to shift the goalposts and declare they were right with their predictions. Current AI certainly doesn’t demonstrate proto-AGI capabilities.

That said, we shouldn’t miss the forest for the trees. We can be skeptical that current The pace of AI progress has been immense and problems that previously seemed difficult (e.g. computer vision classification, or beating top players at Go) have fallen one by one. And AI-skepticism’s have themselves been moving the goalposts in response. I see no reason why composition won’t be the same with time. Indeed, a decade ago machine translation used to struggle to understand the relationships between things, but now seems to be reliable at preserving the compositional relationships post-translation. 2029 is rather optimistic, but AGI does seem to be approaching in the coming few decades.