The author notes how Tensorflow on the M1 isn't useable and hasn't been this whole time, but then uses this as one of the reasons why the M2 needed to be returned? Did he not realize that the same would be true on the M2 as the M1? If he needed fast Tensorflow for his workflow and bought an M2 mac expecting better I'm not sure what to say. (Also who is using TF? every hot thing recently is using PyTorch and while Metal support in pytorch has had struggles at least people are getting things working)

I don't disagree that with the main conclusion, it is a lot of money and for most people it doesn't make sense. If I was spending $2600 I'd be looking at places that do certified refurb M1 Ultra Mac Studios with at least 64gb unified memory, which I think will be going a long way soon to helping future proof a machine for the next few years.