Ask HN: Best prompt to show that AI isn't ready to take over
TL;DR
The HN post is a small Ask HN experiment: it asks for a prompt that makes it obvious AI is not yet reliable enough to take over everything. The concrete example asks Claude for famous cars named after fish, using the Plymouth Barracuda as the reference case. The test targets a familiar weakness: with niche knowledge, models can blend real answers, plausible inventions, and loose category matches.
Nauti's Take
The prompt is simple, but the point lands: many AI demos look strong until someone checks the details. In a list of fish-named cars, one confident false answer is enough to show that language models are still better treated as assistants than autonomous decision-makers.
The stronger version would be a small benchmark of verifiable questions with explicit error tolerance, not one clever prompt.
Briefingshow
These small tests matter because they judge AI on narrow, checkable edge cases rather than polished demo tasks. The key question is not whether the model produces a list, but whether it flags uncertainty, defines the category cleanly, and avoids false positives. That is where everyday reliability shows up.
Sources