问问立正隐私政策
更新于2026年10月15日
问问立正是立正(Yuzheng Sun)做的AI问答工具,包括网页ask.lizheng.ai、lizheng.ai主页上的问答区,以及问问立正的iPhone和安卓App。App里显示的就是同一个网页,适用同一份政策。
简单说:这里像在一场有录像的讲座上举手提问。问题和回答会公开,别人看得到、搜得到;但不记名,我们不知道是谁问的。请只写愿意当众说的话。
公开问答与同意
问问立正用公开的文章、视频和课程文字稿回答问题。你提问时,同意保存和公开使用问题、完整回答与出处,帮助有同样问题的人,并用于改进回答和内容选题。每个问答都会原样展示在「别人在问什么」,有自己的公开页面,能被搜索找到;你也可以把问答分享出去。只写你愿意公开的内容。公开使用以提问时展示且你同意的说明为准。
我们保存什么、为什么
- 提问和回答。你的提问、AI生成的完整回答和所用出处会保存下来,用来改进回答、了解大家关心什么,立正也会从中找写文章、做视频的选题。每个回答过的问题都会和回答一起原样公开在「别人在问什么」,开了「结合我的处境」的也一样。一段对话是一条:标题是它的第一个问题,接着问的问题和回答都在同一个网页里。网页显示问题和回答,不显示你填的处境原文,搜索引擎也能找到;页面上显示大约什么时候问的(精确到5分钟)。回答可能引用处境中的细节。你也可以把自己的问答分享出去。
- 「结合我的处境」。你填写的处境会和提问一起发给AI模型,只用来生成这次回答;我们不保存处境原文,也不单独公开它。回答可能引用其中的细节,并随完整回答保存、在上述范围内公开使用。
- 匿名使用统计。随机生成的浏览器标识和对话标识、这是第几轮、问题字数、问法和入口,用来了解使用情况。
- 每天的次数。浏览器里一个签名的随机标识,用来计算你今天问了几次;同一网络的免费提问总数按网络地址的加密摘要计数。
- 验证Founding身份(可选)。你输入的邮箱用来发送验证码,并向超线性学院社区核验你是不是Founding Member。登录状态加密保存,最长12小时。
- 点赞(可选)。不用登录,按浏览器计数:从浏览器里那个随机标识换算出一个点赞专用的标识,记下它点过哪些公开问答,只用来避免重复计数,不和提问记在一起。为了防止刷赞,同一网络每天的点赞次数按网络地址的加密摘要计数,两天后删除。你点过哪些,也记在这个浏览器里,方便显示。
- 「我的提问」只在你的设备上。在ask.lizheng.ai和App里,你问过的问题和回答存在这台设备的浏览器里,一段对话一条,最多50条,方便回头看、接着问或分享;处境原文不存。这份记录不会发给我们:提问不和账号、邮箱或IP记在一起,我们不知道哪个问题是谁问的,也就没法替你保存或找回。换了设备或浏览器、清除了浏览器数据,就看不到了;也可以在「我的提问」里删掉单条或全部清除。
- 访问统计。页面访问和按钮点击的匿名计数(比如点开了哪一类出处),以及每次打开看了多久、滑到哪里、有没有看到和点开「别人在问什么」、分享了几次、隔几天有没有再来。这些按天合计成数字,用的是随机的浏览器标识和第一次来的日期,不含提问内容。
- 反馈问题(可选)。你点「反馈问题」发送的内容,会连同这个问题、等了多久、出错信息、回答网页的链接(如果有)和设备浏览器类型,用邮件发给立正,用来查问题、修毛病;不包括你填的处境。为了防止刷屏,同一网络每天发了几条按网络地址的加密摘要计数,两天后删除。
我们不做的事
- 提问记录不关联邮箱、账号或IP。登录信息不和提问记在一起,也不交给AI模型。但你输入的文字仍可能包含能识别个人的信息,发送给AI前不会自动去掉;请别填写私密信息。
- 不保存原始IP,不收集位置、通讯录或广告标识,不做跨App跟踪,不出售数据,也不用于广告。
- App只在你选择保存图片时把回答长图写入相册,不读取相册。安卓App下载PDF时把它存进「下载」,不读取其他文件。
谁会处理这些数据
- Builder Space(space.ai-builders.com):运行问答应用,并通过AI网关转发下面两类请求。AI网关只记录模型名、token数、费用估算、时间戳和工具调用次数等用量元数据,不保存问题和回答的正文;这与本产品在Upstash保存问答记录是两回事。
- DeepSeek(官方API,api.deepseek.com):当前使用deepseek-v4-flash生成回答。它接收当前问题、开启后填写的处境、最近最多6轮的问题与回答摘要,以及检索到的公开资料和分析材料。核对或修复回答时,生成的回答也会再次交给它处理。公开问答的主题归类也用同一回答AI,接收问题、完整回答和已有主题,只决定归到哪个主题,不决定是否公开。
- OpenAI:使用text-embedding-3-small匹配公开资料,接收由当前问题、处境以及必要时的上一轮问题组成的检索文本;公开资料的向量也由它生成,用于匹配。它也是回答的备用:DeepSeek在30秒内没有开始写回答,或在写之前出错时,这一次回答改由OpenAI的gpt-5生成,它接收和DeepSeek相同的内容(当前问题、开启后填写的处境、最近最多6轮的问题与回答摘要,以及检索到的公开资料和分析材料)。
- Vercel:网站托管和访问统计。
- Upstash:保存次数、登录状态、问答记录和访问统计。
- Resend:发送验证码邮件,并把「反馈问题」用邮件发给立正。
- Circle(超线性学院社区):核验Founding Member身份。
邮箱、登录信息和账号身份不交给上述AI模型。输入原文在AI处理前不会自动去掉个人信息。OpenAI的API数据规则规定,默认不将API数据用于训练,显式选择共享除外;滥用监控日志通常最多保留30天。DeepSeek通用隐私政策说明,它不覆盖开发者应用中终端用户的数据处理。当前处理链以Builder正式隐私与数据处理说明为准:DeepSeek可能保留输入并用于改进服务或模型,当前没有训练退出机制;本服务没有启用零数据保留。请只提交愿意公开、愿意由这些AI处理的内容。
Builder公布的保护措施包括HTTPS/TLS传输、仅保存密钥哈希和前缀、对数据库限定授权人员和服务的访问,以及按适用法律和处理协议协调删除请求。使用元数据随平台账号保留,账号删除时清除,法定保留除外;平台不长期保存的请求内容仍可能由下游保留。我们的删除请求会在适用法律和各处理方政策范围内转交与协调,不承诺删除他人已保存的公开副本。
App中的AI同意
在App里,第一次AI提问前会单独展示公开用途、发送的数据、Builder Space及实际AI接收方。只有点「同意AI处理与公开使用」才会发送;点「不同意」不发送,也不扣提问次数,仍可浏览公开问答和出处。披露、接收方或回答模型变化时会重新征求同意。
可以在「说明」里点「撤回AI数据发送同意」。撤回后,新提问和重新生成回答都需要再次同意。撤回不收回已经发出的内容,也不自动删除已有记录;如需删除,请按下文联系我们。
保存多久
- 问答记录保存到立正删除为止;公开的问答随时可以撤下,撤下后它的网页大约十分钟内失效,搜索引擎下次来抓取时会把它移除。
- 问答记录每天另存一份备份,留30天。删除的问答在之前的备份里最多再留30天,随备份一起删除。
- 次数记录两天后自动删除;验证码10分钟内有效;登录状态最长12小时;Founding身份的核验结果最多保存30天,这样不用每次登录都重新核验。
- 按天的访问统计400天后删除;浏览器里记的第一次来的日期400天后过期。
- 「反馈问题」的邮件留在立正的邮箱里,到他删除为止;反馈的每日计数两天后删除。
- 点赞记录随那条公开问答保存,问答删除时一并删除。浏览器里记的「我的提问」和点过的赞,留到你删除或清除浏览器数据为止。
你可以
- 不登录直接使用,每天3次。
- 只写你愿意交给AI处理并公开使用的内容。「结合我的处境」默认关闭。
- 要求删除:记录不关联身份,我们没法按人查找。如果想删除自己问过的问题,或撤下公开的问答,请写信告诉我们大致时间和问题内容(分享过的,附上链接就行),我们找到后删除本产品记录、撤下页面。需要删除已发给AI的数据时,请在同一封信里说明,我们会向Builder协调请求,按适用法律和对应服务商规则处理;已经公开的内容可能被他人保存。
儿童
问问立正不面向13岁以下的儿童。
变更与联系
政策有变化时,我们会更新本页和上面的日期。有任何问题,请写信到yz@superlinear.academy。
Privacy Policy
Updated October 15, 2026
问问立正 (Ask Lizheng) is an AI question-and-answer tool made by Yuzheng Sun (立正). It covers the website ask.lizheng.ai, the Q&A section on the lizheng.ai homepage, and the 问问立正 apps for iPhone and Android. The apps show the same web page, so the same policy applies.
In short: asking here is like raising your hand at a recorded talk. Questions and answers are public, others can see them and search engines can find them, but no names are attached and we don’t know who asked. Write only what you’d say in front of everyone.
Public Q&A and permission
Ask Lizheng answers questions using public articles, videos and course transcripts. Asking means agreeing to save and use the question, complete answer and sources publicly, to help others, improve answers and find content topics. Every question and answer is shown as asked in “别人在问什么”, on its own public page that search engines can find, and you can share it. Write only what you are willing to make public. Public use follows the disclosure shown and accepted when you ask.
What we keep, and why
- Questions and answers. Your question, the complete AI answer and the sources it used are kept to improve answers, to learn what people care about, and to find topics for essays and videos. Every answered question is published as asked, with its answer, in “别人在问什么” (What others ask), including questions asked with “结合我的处境”. A conversation is one entry: it is titled by its first question, and its follow-up questions and their answers are on the same web page. The page shows the questions and answers, without the situation you wrote, and search engines can show it, with roughly when it was asked (to five minutes). The answer may quote details from your situation. You can also share your own question.
- Your situation (optional). If you turn on “结合我的处境”, what you write is sent to the AI model with your question, only to produce this answer. We don't keep the situation text or publish it separately. The answer may quote its details, which are saved with the complete answer and used publicly within the scope above.
- Anonymous usage statistics. Random browser and conversation identifiers, the turn number, question length, the way of asking and the entry point.
- Daily limit. A signed random identifier in your browser counts today's questions; free questions from the same network are counted by a keyed hash of the network address.
- Founding Member verification (optional). The email you enter receives a one-time code and is checked against the Superlinear Academy community for Founding Member status. The signed-in state is kept encrypted for up to 12 hours.
- Likes (optional). No sign-in needed; likes count by browser. A like-only identifier derived from your browser's random identifier records which published answers it liked, only to avoid double counting, and is never stored with questions. To prevent like stuffing, likes from the same network each day are counted by a keyed hash of the network address, deleted after two days. Your browser also remembers which ones you liked, to show them.
- “我的提问” (My questions) stays on your device. On ask.lizheng.ai and in the app, the questions you asked and their answers are kept in this device's browser, one entry per conversation, up to 50, so you can reread, follow up or share them; the situation text is not kept. This list is never sent to us: questions aren't linked to an account, email or IP address, so we don't know who asked what and can't keep or restore it for you. Another device or browser, or clearing browser data, starts empty; you can also delete one or all in “我的提问”.
- Visit statistics. Anonymous counts of page views and button taps (for example, which kind of source was opened), how long each visit lasted, how far the page was scrolled, whether “别人在问什么” was seen or opened, how often answers were shared, and whether the browser came back on a later day. They are totalled by day using a random browser identifier and the date of the first visit, never the question text.
- Problem reports (optional). When you use “反馈问题” (Report a problem), what you write is emailed to Yuzheng together with that question, how long it waited, any error, the link to the answer's page (if there is one) and your device and browser type, so he can find and fix the problem. Your situation is not included. To prevent flooding, reports from the same network each day are counted by a keyed hash of the network address, deleted after two days.
What we don't do
- Question records are not linked to an email, account or IP address. Sign-in details are never stored with questions or sent to the AI model. However, the text you enter may identify you and is not automatically redacted before AI processing. Please do not enter private information.
- We don't keep raw IP addresses, collect location, contacts or advertising identifiers, track you across apps, sell data or use it for advertising.
- The app writes an answer image to your photo library only when you choose to save it, and never reads your photos. When you download a PDF, the Android app saves it to Downloads and reads no other files.
Who processes the data
- Builder Space (space.ai-builders.com) hosts the Q&A application and forwards the two types of requests below through its AI gateway. The gateway keeps usage metadata such as model names, token counts, cost estimates, timestamps and tool-call counts, without storing question or answer bodies. This is separate from this product's Q&A archive in Upstash.
- DeepSeek (official API, api.deepseek.com) currently generates answers using deepseek-v4-flash. It receives the current question, enabled situation, questions and answer summaries from up to six recent turns, and retrieved public sources and analysis material. Generated answers may be sent back for validation or repair. The same answering model also files each public question under a topic, receiving the question, complete answer and existing topics; it does not decide whether a question is published.
- OpenAI matches public sources using text-embedding-3-small. It receives search text composed of the current question, situation and, when needed, the previous question, and it also generates vectors for the public source material used in matching. It is also the backup for answers: when DeepSeek has not started writing within 30 seconds, or fails before it writes, that answer is generated by OpenAI's gpt-5 instead, which receives the same content as DeepSeek (the current question, enabled situation, questions and answer summaries from up to six recent turns, and retrieved public sources and analysis material).
- Vercel hosts the website and counts visits.
- Upstash stores daily counts, sign-in state, question records and visit statistics.
- Resend sends verification emails and delivers problem reports to Yuzheng by email.
- Circle (the Superlinear Academy community) confirms Founding Member status.
Email, sign-in details and account identities are not sent to these AI models. Input is not automatically redacted before AI processing. OpenAI's API data rules exclude API data from training by default unless explicitly shared; abuse-monitoring logs are normally retained for up to 30 days. DeepSeek's general privacy policy says it does not cover end-user data processing in developers' applications. The current route is documented in Builder’s published Privacy and Data Processing Policy: DeepSeek may retain inputs and use them to improve services or models, with no training opt-out on this route. Zero data retention is not enabled. Submit only content you are willing to make public and have these AI services process.
Builder documents HTTPS/TLS transmission, hashed keys and prefixes, restricted database access, and coordination of deletion requests under applicable law and processor agreements. Usage metadata remains for the platform account’s lifetime and is removed on account deletion, except for legally required records. Downstream services may retain request content. We pass on and coordinate deletion requests within applicable law and each processor’s policy; copies of public content saved by others cannot be recalled.
AI permission in the app
Before the first AI question in the app, a separate dialog lists the public uses, data, Builder Space and actual receiving AI services. Data is sent only after you tap “同意AI处理与公开使用” (Agree to AI processing and public use). “不同意” (Do not agree) sends nothing and does not use your daily allowance; you can still browse public answers and sources. Changes to the disclosure, recipients or answer model require fresh permission.
In “说明” (About), tap “撤回AI数据发送同意” to withdraw permission. New questions and answer retries then require permission again. Withdrawal cannot recall data already sent and does not automatically delete existing records. Contact us as described below to request deletion.
How long we keep it
- Question records are kept until the owner deletes them; published questions can be withdrawn at any time. A withdrawn question's page stops working within about ten minutes, and search engines drop it when they next visit.
- A backup copy of the records is made each day and kept for 30 days. A deleted question stays in earlier copies for at most 30 more days, then leaves with them.
- Daily counts are deleted after two days; verification codes expire after 10 minutes; sign-in lasts at most 12 hours; a Founding Member check result is kept up to 30 days, so you are not checked again at every sign-in.
- Visit statistics by day are deleted after 400 days; the first-visit date kept in your browser expires after 400 days.
- Problem reports stay in Yuzheng's mailbox until he deletes them; the daily report counts are deleted after two days.
- Likes are kept with their published question and deleted with it. “我的提问” and the likes your browser remembers stay until you delete them or clear browser data.
Your choices
- Use it without signing in, 3 answers a day.
- Write only what you are willing to have processed by AI and used publicly. “结合我的处境” is off by default.
- Deletion: records aren't linked to identities, so we can't look them up by person. Email us the approximate time and wording of your question (or its shared link), and we will find and delete the product's record and take down its page. To request deletion of data already sent to AI, include this in the same email and we will coordinate with Builder under applicable law and the receiving service’s rules. Others may have saved content that was already public.
Children
问问立正 is not directed to children under 13.
Changes and contact
We will update this page and its date when the policy changes. Questions: yz@superlinear.academy.