Is it just me, or is there a yawning gap between the claims of the AI vendors and the lived reality of AI users?

Today I simply asked my AI to read through a set of documents and point out any inconsistencies. It overlooked some, hallucinated some, and suggested convoluted multi-paragraph rewrites where I could just change one sentence.

It does apologize profusely, but I would rather have it just do what I asked it to. In real life, we are a long way from what the vendors say they already have, and light years from the AI apocalypse.