V2EX 用户用 ZooKeeper 一致性问题测试 AI，发现回答不一
英文摘要
A V2EX user posed a distributed systems question: after sequentially writing keys 1 through 5 in ZooKeeper, can three clients simultaneously observe the sequences [1-5], [2-5], and [3-5]? Different AI assistants gave contradictory answers. The user’s own reasoning, based on ZooKeeper’s sequential consistency guarantees, argues it is impossible because seeing a later write implies having seen all earlier writes. The user then fed their reasoning back to the AIs, which then agreed, highlighting that LLMs can flip opinions and lack reliable reasoning on nuanced technical topics. The post is framed as an informal evaluation of AI judgment and intelligence.
中文摘要
V2EX 用户提出一个分布式系统问题：在 ZooKeeper 中顺序写入键 1 到 5 后，三个客户端能否同时读到 [1-5]、[2-5] 和 [3-5]？不同 AI 助手给出了矛盾的答案。用户基于 ZooKeeper 的顺序一致性保证，认为不可能出现这种读模式，因为读到后续写入意味着必然读到了之前的所有写入。随后用户将这一推理发给 AI，AI 们又转而认同，暴露出大语言模型立场摇摆、在技术细节上缺乏可靠推理。该帖子被当作对 AI 判断力和智力的一次非正式评测。
关键要点
A V2EX user asked a specific ZooKeeper consistency question about client read views after sequential writes.
V2EX 用户提出关于 ZooKeeper 顺序写入后客户端读视图的一致性问题。
The user consulted multiple AI assistants, and they gave inconsistent answers, with some suggesting the described read pattern is possible.
用户询问多个 AI 助手，它们给出了相互矛盾的答案，部分认为所述读模式可能发生。
Based on ZooKeeper’s sequential consistency model, the user argued the pattern [1-5], [2-5], [3-5] cannot coexist because a reader that sees 2 must have seen 1, making a prefix-only read impossible.
根据 ZooKeeper 的顺序一致性模型，用户论证 [1-5]、[2-5]、[3-5] 不可能同时出现，因为读到 2 的客户端必须也读到了 1，不存在仅读后缀的情况。
When the user presented their reasoning to the AIs, the models agreed, demonstrating a tendency to conform rather than provide robust technical judgment.
当用户将推理提供给 AI 后，模型们又表示认可，表现出迎合而非坚实技术判断的倾向。
The user viewed this as a real-world test of AI intelligence and reliability on distributed systems topics.
用户将此视为对 AI 在分布式系统话题上智力与可靠性的一次实际测试。