About this experiment
AI can do remarkable things. It also has peculiar gaps, and people keep finding ambitious ways to make those gaps everyone else’s problem.
This is a collection of specific examples: the chatbot that takes a joke literally, the company that automates the wrong task, the impressive demo that probably needed another test.
What makes a good entry?
Something concrete happened. There is a source to check. And the gap between the pitch and the outcome is worth a closer look. Harm alone does not make a story a fit; the interest is in the mistake, the absurdity, and the human decisions around it.
How to read the sources
Primary source directly documents the claim. Reported relies on a publication’s reporting. Firsthand account is a personal observation with its limits stated beside the entry. Those labels describe the available evidence.
Each entry’s notes explain what the AI did, what the people contributed, and any relevant caveats. The Amazon example is a personal account; the linked announcement confirms the feature exists, not that the exchange happened.
Who edits this?
This is an early prototype. AI helps discover and draft candidates; an editor decides which ones belong here. The first research batch is preserved for testing the review process. Unreviewed drafts are excluded from the reader.
Corrections and submissions
A public contact channel will be added before a wider launch. For now, corrections and withdrawals are recorded in the project’s source history.