Hacker News 中文摘要

RSS订阅

Codex开始加密子代理提示 -- Codex starts encrypting sub-agent prompts

文章摘要

我们认真阅读每一条反馈,并高度重视您的意见。如需联系,请提供您的邮箱地址。

文章总结

我们认真阅读每一份反馈,并高度重视您的意见。如需联系,请附上您的邮箱地址。

评论总结

根据评论内容,总结如下:

主要观点与论据:

  1. 标题误导性(多位用户指出):原始标题“Codex starts encrypting prompts, uses ciphertext for inference instead”易被误解为同态加密推理,实际仅针对子代理通信加密,推理仍在明文进行。

    • 引用:xnorswap: "HN Title is very misleading, it makes it sound like inference is being done directly on ciphertext"(评分None)
    • 引用:mpeg: "they're not using ciphertext for inference – they're passing ciphertext around in cases where an agent calls into another agent"(评分None)
  2. 反制蒸馏与非法转售(部分用户支持):加密主要针对中国黑市转售商和模型蒸馏行为,保护商业利益。

    • 引用:londons_explore: "mostly to frustrate efforts to proxy large numbers of user requests... to train competitor models"(评分None)
    • 引用:next_xibalba: "obviously a countermeasure against distillers, illicit resellers... Chinese black market has become a serious threat"(评分None)
  3. 功能范围有限:加密仅适用于multi_agent_v2实验性功能,子代理输出仍为明文,且不影响普通GPT订阅使用。

    • 引用:fortuitous-frog: "only for parent -> subagent spawns/messages, and only for the multiagentv2 feature"(评分None)
    • 引用:pshirshov: "If not - that doesn't matter much, codex cli is a remarkably unremarkable harness"(评分None)
  4. 对开源与用户的影响:部分用户质疑此举与开源承诺的矛盾,以及非Codex工具链的兼容性。

    • 引用:pradeep1177: "Then why to even keep codex open source?"(评分None)
    • 引用:resonious: "non-Codex harnesses get a little bit worse? ...what's so special about their subagents system"(评分None)

平衡性说明: 评论整体对加密措施持中性偏负面态度,主要批评标题误导性,同时认可其反制蒸馏的合理性。未出现强烈支持或反对的极端观点。