我做出來的 AI 根本反向操作：不是憑空生成答案，而是一個一個刪除錯誤選項——這樣準確度爆表，來問我啦 [P]

For the past few months I’ve been building POEM (Process Of Elimination Master) — a standalone AI architecture that reaches answers by progressively eliminating impossibilities rather than generating possibilities. No LLM dependency. Instead of predicting tokens, POEM classifies the question, eliminates wrong categories, then searches a structured knowledge base. The answer is what cannot be eliminated. One of the core motivations is energy. LLMs run a full billion-parameter forward pass for e