The problem is that he relies on a dubious Acemoglu estimate, without realizing that at best it's temporary. AI will be better than humans in doing tasks (the qualifiers don't matter in the aggregate). Any jobs then would be bullshit jobs, and everyone will know it.
LMs are literally trained on Reddit. The idea they are "super-intelligent" is anthropomorphism and marketing. So far no independent research has seen AI be better at anything than humans. And that's unlikely to change with larger NNs.