Can LLMs identify 16 cards in 45 bit-queries?

A shiny new post about using an artificial intelligence chatbot to attack a fiendish card puzzle should have been a victory lap for math-meets-machine hype. Instead, the real spectacle broke out in the replies, where readers instantly turned into fact-checking detectives, skeptics, and chaos gremlins. The puzzle sounds simple enough for non-math people to follow: hidden cards, yes-or-no questions, and the challenge of figuring everything out in fewer moves than the obvious brute-force method. But the moment the article claimed the smaller version could be solved in under 5 questions, the comment section smelled blood.

One reader flatly declared the simulator looked busted, accusing it of spitting out the same weird answer pattern and dropping the insult “vibe slop” like a gauntlet. Another came in with the cooler-but-deadlier energy of a person doing the arithmetic on a napkin and saying, hold on, there are 24 possible arrangements here, so why is this being sold as always solvable in under five questions? Translation for everyone at home: the community wasn’t just debating the big AI achievement — they were questioning whether the demo itself was playing fair.

That’s what makes this such delicious internet drama. On one side: wonder about whether large language models, or AI text systems, can help crack real unsolved math problems. On the other: commenters basically yelling, “Before we crown the robot, can someone fix the toy?” It’s part math flex, part bug hunt, part public trust exercise — and the crowd clearly thinks the details matter as much as the headline.