正在加载陈远的个人网站

语音意识流与 AI 追问:把想法说清楚

用语音意识流、AI 追问和复述校对,逐步澄清未成形的想法。

AI 应用研发效能

来源摘录

分享一个 Karpathy 的用法,我看完立刻加进了自己的工作流 我本来就重度依赖语音输入,跟 AI 对话从来不打字,想法有多细说多细,但他这个思路更好: 1、/voice,对着 LLM 意识流乱说 10 分钟,想到哪说到哪 2、说不下去就让 AI 反过来采访你,几轮问答把想法全逼出来 3、最后让它把你的想法复述一遍,那个版本经常比你脑子里的还清楚 聊几轮下来,想不清楚的事,说着说着就清楚了,这可能是 /voice 被低估的用法

One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.

— @karpathy(Andrej Karpathy)的引用原文

深度解析

是什么 这是一种以语音为入口的 LLM 思考整理工作流。用户先用约十分钟连续表达未经整理的想法,再在卡住时让 AI 通过几轮访谈主动追问,最后要求模型复述整体意图。

核心设计 它把交互拆成“低摩擦输入—主动澄清—结构化回声”三段。语音降低输入成本,使用户能够提供更多上下文;采访模式让模型从被动响应者变成提问者;复述则提供一个可校对的外部认知镜像。整体节奏是先发散、后收敛。

关键洞察 长而不完整的表达不一定是噪声,足够上下文可能帮助模型重建用户意图。复述的价值也不只是摘要,而是暴露隐藏假设,让用户发现自己真正想解决的问题。这种模式可借鉴到需求梳理、写作构思和 Agent 工作流中。

边界与风险 语音识别错误、模型过度脑补、追问方向偏置或过长上下文,都可能制造虚假的清晰感。涉及决策、隐私或高风险事项时,应保留原始转写,逐条核对复述内容,并区分用户原意、模型推断与最终结论。