LLM 缺乏 AlphaGo 級別的推理能力

ten years after alphago defeated go champion lee sedol, today's ai systems still lack the genuine reasoning capabilities that made alphago's win possible. alphago's search machinery, which constructed and searched a game tree with thousands of branches, allowed it to make creative choices like move 37 in game two. in contrast, large language models rely on fast, associative thinking, generating tokens one by one. while recent models have introduced intermediate steps to decompose problems, these are still produced by the same next-token prediction mechanism, not a separate reasoning mechanism.