If you give a stapler the gorilla test it fails. If you give a stapler the ability to see gorillas then give it the gorilla test it succeeds.

The initial blog post effectively proved the former. I appreciate that this post proves the latter through demonstrating how to add the ability to see gorillas to the stapler. But I think concluding that either proves a quality of staplers rather than our expectations of them is unwise.

Which is more or less the conclusion this post comes to, I suppose; but I think the source of the dissonance in the first place comes from thinking about AI as "humans minus [x] capabilities" rather than purpose-built machines. And I think until that fundamental (mis)understanding is clarified people will still be making poor assumptions about what AI can and cannot do.