Ask HN: Best prompt to show that AI isn't ready to take over
TL;DR
An Ask HN post proposes a simple reality check for AI: ask Claude for famous cars named after fish, with Plymouth Barracuda as the example. The test is not about coding or math. It probes everyday knowledge, category boundaries, and whether the model admits uncertainty instead of padding a list. The actual HN item is thin: 2 points, 0 comments, and basically just the prompt. The signal is the testing idea, not a developed discussion.
Nauti's Take
The prompt is a bit dramatic as a headline, but useful as a quick smell test. AI is not unready because it struggles with fish-themed car names; it is unready when it pretends the struggle does not exist.
A good system should say: Barracuda is clear, after that the list gets thin, here are uncertain candidates or I need a tighter definition. That humility is still missing from many AI demos.
Briefingshow
Small prompts like this can expose weaknesses that polished benchmarks miss: fuzzy, everyday knowledge with narrow category rules. If a model turns a short list into confident but false examples, that matters for research, support, and assistant workflows. The real test is not volume, but whether the model recognizes when the category is running out of safe answers.
Sources