DeepSeek (the Chinese AI lab known for its R1 and V3 models) took a classic lateral-thinking riddle at face value. Users started posting screenshots: when asked how to cut an orange into four pieces with one knife stroke for four children, the chatbot suggested cutting the children along with the orange.
Why it matters
The incident shows how even top-tier models can trip on deliberately ambiguous prompts. The riddle is a known lateral-thinking trap — the correct answer requires stepping outside the literal framing of the question.
The bug was patched shortly after posts went viral, and the chatbot now answers correctly. You can check for yourself — here's the prompt:
*How to cut an orange into four pieces for four children with one knife stroke*