深港两地的每一天，
一件一件用 AI 做对
Everyday life across two cities,
one thing at a time, done right.
从深港居民重复次数最多的那几件事开始Starting from what people here repeat most. Starting where two cities meet, every day.
深港人工智能应用创新中心。我们的选题只看三个乘数：一件事每天重复多少次、两地的规则差多少、做错了用户能不能当场看出来。三项都高的才进清单 —— 过关通勤、两地消费、就医配药、升学查询。至于模型该跑在设备上还是云端，由这件事本身的约束决定，不由我们的技术偏好决定。 We choose what to work on by three multipliers: how many times a day it repeats, how far the rules diverge between the two cities, and whether a mistake is visible to the user on the spot. Only when all three are high does something make the list — crossing the border, paying on either side, filling a prescription, checking a school rule. Whether the model runs on the device or in the cloud is then decided by the constraints of that particular thing, not by our preference in technology.
从一天里重复最多次的事情出发Starting from what a day repeats most
深港人工智能应用创新中心。我们把深港居民使用频次最高的生活场景排成一张清单，然后从最上面那几件开始做。The Centre works on putting frontier AI to use, and it starts where people on both sides of the border actually spend their days.
我们不从行业大词出发，而从一件具体的事出发：一个人早上要过关、中午要点餐、下午要配药、晚上替孩子查升学规则。这些事每天发生上百万次，两地的规则、语言与证件却不一样 —— 通用模型在这里出的错，普通人一眼就能看见。这既是难处，也是最好的产品起点。We do not begin with an industry category; we begin with something concrete. Someone crosses the border in the morning, orders lunch, fills a prescription in the afternoon, and looks up a school rule that evening. These things happen millions of times a day, and the rules, the language and the paperwork differ on either side — so when a general model gets it wrong, an ordinary person sees it immediately. That is the difficulty, and it is also the best place to start a product.
从研究、适配到部署，我们关注的始终是同一件事 —— 把前沿能力，变成一个人今天就能用上的东西。From research and adaptation to deployment, we focus on one thing throughout — turning frontier capability into something a person can use today.
频次优先Frequency first
How often it repeats两地差异Two sets of rules
Where the rules diverge现场交付Delivered on site
Someone sits next to the user我们专注的三件事Three things we focus on
我们不是一家以某项技术命名的机构。顺序是反过来的：先确定哪件事值得做，再由它的约束决定该用什么技术、跑在哪里。以下三件事，就是这个顺序本身。We are not an organisation named after a technology. The order runs the other way: first decide which everyday thing is worth doing, then let its constraints decide what technology to use and where it runs. The three items below are that order.
选题 · 只做最高频的那几件Picking · Only the most frequent fewFrequency × divergence × visibility
按三个乘数排序：每天重复多少次、两地规则差多少、做错了能不能被当场看见。三项都高的才进清单。听起来很重要、但一个人一年只遇到一次的事，我们不做。Ranked by three multipliers: how often it repeats, how far the two cities' rules diverge, and whether an error is visible on the spot. Only when all three are high does it make the list. Things that sound important but happen to a person once a year, we leave alone.
端还是云 · 由约束决定On device or in the cloud · The constraint decidesWhere it runs is not a preference
涉及个人信息、又常发生在弱网下的（配药、证件文书），推理只能留在设备上，这是前提不是加分；需要时效与核实的（升学规则、班次票务），必须联网检索。真正的难题是那条切换线画在哪里 —— 这也是我们最想和高校一起做的题目之一。Where personal data is involved and the signal is often weak — prescriptions, documents — inference has to stay on the device; that is a precondition, not a bonus. Where currency and verification matter — school rules, timetables — it has to go online and retrieve. The hard question is where the line between them falls, and it is one of the topics we most want to work on with a university.
交付 · 有人坐在用户旁边Delivery · Someone sits next to the userThe forward-deployed engineer
FDE（前向部署工程师）把一句「不好用」翻译成可度量的问题，在算力、时延、电量与隐私的预算里做取舍，再把方法送到日常能被使用的状态。论文之后、产品之前的这一段，目前既缺人，也缺研究。A forward-deployed engineer turns "it doesn't work well" into a measurable problem, trades off within the budgets of compute, latency, battery and privacy, and carries the method to the point where it survives daily use. That stretch — after the paper, before the product — is short of people and short of research.
深港 AI 与生活场景生态的连接节点A connecting node between AI and everyday life across SZ–HK
中心不只做技术，更串起生态各方 —— 让高校的前沿研究、深圳的端侧算力、两地的真实用户与国际资本，在同一件具体的事情上对得上。Beyond technology, the Centre links the players — university research, Shenzhen's edge compute, real users on both sides and international capital — around one concrete thing at a time.
我们出现场，题目由你来提We open the site. You bring the question.
多数产学研合作是企业把题目写好、交给实验室去做 —— 那样得到的通常是一份外包，不是研究。我们反过来：把深港两地高频生活场景的现场开放出来，包括端侧设备、真实用户与两地不同的规则，题目请老师和学生自己提。中心这一侧的主线只有一条 —— FDE（前向部署工程师）：有人坐在用户旁边，把一句「不好用」翻译成可度量的问题，再把方法送到日常能用的状态。这条线上没被回答的问题非常多，而且它们只有在现场才问得出来。Most industry–academia arrangements have the company write the question and hand it to a lab — which usually produces outsourcing rather than research. We work the other way round: we open the site — edge devices, real users on both sides of the border, and two different sets of rules — and the question comes from you. On our side there is one through-line: the forward-deployed engineer. Someone sits next to the user, turns “it doesn’t work well” into a measurable problem, and carries the method to the point where it holds up in daily use. There are a great many unanswered questions along that line, and they can only be asked on site.
学术的那一半留在高校，现场的那一半由我们承担The academic half stays with the university. We carry the site.
中心不授学位、不占招生指标、不参与录取。学籍、导师关系、学术评价与毕业标准，全部留在高校一侧。题目由高校导师与学生提出，中心不干预学术结论，只负责把现场、设备与用户准备好。The Centre does not award degrees, does not hold admission quotas, and takes no part in admissions decisions. Enrolment, supervision, academic assessment and graduation standards all remain with the university. The question comes from the supervisor and the student; we do not interfere with academic conclusions — we prepare the site, the devices and the users.
目前正在与内地及香港的高校实验室推进首批共建，以单实验室、单课题、单学生为起步单元；一份一页的提案就可以开始谈。We are establishing our first partnerships with labs in the mainland and Hong Kong, starting with one lab, one topic, one student. A one-page proposal is enough to start the conversation.
- 现场Site深港两地高频生活场景的真实使用环境，不是模拟数据集。Real everyday settings on both sides of the border — not a simulated benchmark.
- 设备Devices端侧设备与国产 NPU 平台，以及工位与算力。Edge devices and domestic NPU platforms, plus workspace and compute.
- 用户Users愿意持续参与测试的两地真实用户 —— 小规模、可重复、会当面告诉你哪里不好用。Real users on both sides who will keep testing — small in number, repeatable, and blunt about what does not work.
- 带教Mentorship一位真正在现场交付过的企业导师，与学术导师并行；需要的话，可以用短期共训／训练营的形式集中带一段。An industry supervisor who has actually delivered on site, working alongside the academic supervisor.
主线是 FDE：把研究做成有人愿意每天用的东西The through-line is the FDE: making research into something people will use every day
FDE，前向部署工程师 —— 坐在用户那一侧，把问题定义出来，再把方法送到能被使用的状态。它不替代学术训练，而是补上论文之后、产品之前的那一段；这一段目前既缺人，也缺研究。The forward-deployed engineer sits on the user's side of the problem, frames it, and carries a method all the way to the point where it can be used. It does not replace academic training; it covers the stretch after the paper and before the product — a stretch short on both people and research.
用户说出口的是「不好用」。第一项工作是把它翻译成可度量的东西：哪一类输入、多少时延、错到什么程度算错、谁来判定。这一步做错，后面所有优化都在解一道假题。What the user says is “it doesn’t work well.” The first job is to translate that into something measurable: which inputs, what latency, how wrong counts as wrong, and who decides. Get this wrong and every optimisation afterwards solves a fake problem.
算力、时延、内存、电量、隐私、离线可用性，每一项都是预算。方法的优劣要在这几条线同时成立的位置上比较，而不是在资源无限的假设下比较 —— 这正是公开榜单给不出答案的地方。Compute, latency, memory, battery, privacy, offline availability — each is a budget. Methods have to be compared where all of them hold at once, not under an assumption of unlimited resources. This is precisely where public leaderboards stop being informative.
能装到设备上、有权限与日志、错了有兜底、结果可复现。这是一项研究成果在真实用户手上活过三个月的最低条件，也是学生简历上最难伪造的一段。It installs on the device, has permissions and logs, falls back gracefully when it fails, and reproduces. That is the minimum for a research result to survive three months in real users' hands — and the hardest line on a student's CV to fake.
我们能开放的现场The sites we can open
以下不是课题，是现场。每一条后面都躺著一堆还没人回答的问题 —— 哪一个值得做成课题，我们想听你的判断。What follows are not topics; they are sites. Behind each sits a pile of unanswered questions. Which of them is worth turning into research — that judgement is what we want from you.
-
过关与通勤Crossing the border, and getting to work
每天数十万人往返两地。最需要判断的那几分钟，往往正好在弱网、换网、换卡的当口；口岸、接驳、票务的信息分散在互不相通的系统里。一个只能在有网时工作的助理，在这里等于没有。Hundreds of thousands cross daily, and the minutes when you most need an answer are exactly the minutes of weak signal, network handover and a switched SIM. Checkpoint, transit and ticketing information sits in systems that do not talk to each other. An assistant that only works online is no assistant here.
-
两地消费与支付Spending and paying on both sides
同一件事，在两边有两套正确答案：支付方式、兑换、单据、报销、退税。模型不但要知道规则，还要知道自己此刻站在哪一边 —— 而它常常不知道。The same task has two correct answers depending on which side you are standing on: payment method, exchange, receipts, reimbursement, tax refund. The model has to know the rules and also know which side it is currently on — and it usually does not.
-
饮食与点单Eating out and ordering
繁简混排的菜单、粤语与普通话夹杂的口语、手写价目与拍糊的照片。这是一个对端侧小模型极不友好、对用户却极其日常的多语种理解问题。Menus mixing traditional and simplified characters, speech switching between Cantonese and Mandarin, handwritten prices and blurry photographs. A multilingual understanding problem that is brutal for a small on-device model and utterly ordinary for the person holding the phone.
-
就医与配药Clinics and prescriptions
两地的药名、剂量单位与处方规则不同，而用户要的往往只是一次可靠的对照与提醒。这里没有任何讨价还价的空间：个人信息只能留在端侧，错了要能被发现，讲不清楚就必须说「去问医生」。Drug names, dosage units and prescription rules differ across the border, and what the user usually needs is one reliable cross-check and a reminder. There is no room for negotiation here: the data stays on the device, errors have to be detectable, and when the answer is not clear the honest output is “ask a doctor.”
-
升学与教育Schooling and education
两地学制、插班与报考规则变动频繁，家长面对的是散落各处、时效性极强、且错一步代价很大的信息。这是一个检索与核实远比生成困难的问题。School systems, mid-year transfer and application rules change often; parents face information that is scattered, time-sensitive, and costly to get wrong by one step. Here retrieval and verification are far harder than generation.
-
语言与文书Language and paperwork
表格、证件、住址、公文用语，在两地有各自的写法与惯例。翻译不是难点，「在这一边该怎么填才会被受理」才是。Forms, identity documents, addresses and official phrasing follow their own conventions on each side. Translation is not the hard part; knowing how it has to be filled in on this side to be accepted is.
课题可以是两类Two kinds of topic
同一个现场，可以问两种问题。两类我们都接，也欢迎一位老师带著两个学生各做一类。The same site supports two kinds of question. We take both, and we are glad to see one supervisor put two students on one each.
在某一个高频场景里，把一个具体的技术问题做到能用：弱网下的离线理解、多语种混排的识别、跨两套规则的核实与对照、端侧小模型的错误如何被发现。成果是方法，也是论文。Take one concrete technical problem inside one high-frequency scenario and carry it to usable: offline understanding on a weak signal, mixed-script recognition, verification across two rule sets, or how a small on-device model's errors get caught. The output is a method, and a paper.
把「怎么送到能用」当作研究对象：问题该如何定义与度量、端与云的切换线画在哪里、日常场景的评测集怎么建、交付之后如何观察与回收。这一段几乎没有系统性的研究，却决定了前一类成果能不能活下来。Treat "how it gets to usable" as the object of study: how a problem should be defined and measured, where the line between device and cloud falls, how an evaluation set for everyday use is built, and how a deployment is observed and learned from afterwards. Barely any of this is studied systematically, yet it decides whether the first kind of result survives.
提案怎么写：一页纸，说清三件事Writing a proposal: one page, three things
不需要立项书格式，不需要先有结果。老师、学生都可以直接来信；同一个现场我们欢迎不同的提法。No formal template, no preliminary results needed. Supervisors and students are equally welcome to write; different takes on the same site are welcome too.
具体到一个人、一个动作、一次失败。「跨境场景的体验优化」不算，「在口岸弱网下，离线助理答不出换乘」算。Specific to one person, one action, one failure. “Improving the cross-border experience” does not count; “on a weak signal at the checkpoint, the offline assistant cannot answer the transfer question” does.
什么指标、在什么设备上、多少样本、跟谁比。允许粗糙，但要能被别人重做一遍。Which metric, on which device, how many samples, compared against what. Rough is fine; reproducible by someone else is not optional.
设备、场景、用户、工位、算力，还是一位在现场待过的人。说清楚了，我们才知道自己接不接得住。Devices, the setting, users, a desk, compute — or a person who has spent time on site. Say it plainly, so we can tell you honestly whether we can provide it.
两周内回复。合适的话，下一步是一起去一次现场，而不是先签一份协议。We reply within two weeks. If it fits, the next step is a visit to the site together — not a signed agreement.
隐私Privacy
生活场景涉及个人信息，因此一律最小化收集、优先端侧处理、参与者可随时撤回。研究开始前先过高校伦理审查与中心的隐私评估；涉及跨境的个人信息处理，按两地各自的规则分别执行。Everyday scenarios involve personal data, so collection is minimised, processing happens on the device wherever possible, and participants can withdraw at any time. Work begins only after university ethics review and the Centre's own privacy assessment; anything crossing the border follows each jurisdiction's rules separately.
发表Publication
方法与结论经脱敏审查后可正常发表。审查范围仅限参与者身份与个人信息，窗口 15 个工作日，无异议即视为通过；中心不以审查为由干预学术结论。Methods and conclusions are publishable once desensitisation review is complete. Review covers participant identity and personal data only, with a 15-working-day window; no response means cleared. Review is never a lever over academic conclusions.
权属IP
学生的学术成果与学位论文权利归学生与高校。中心不主张对通用方法学的独占；仅当技术与中心既有产品或特定场景直接结合、形成可专利成果时，双方就共有与实施许可另行约定。Academic outputs and thesis rights belong to the student and the university. The Centre claims no exclusivity over general methodology; only where the work combines directly with an existing product or a specific setting to form a patentable result do we agree separate terms on joint ownership and licensing.
学生毕业与发表，不受商业条款掣肘。这是前提，不是让步。A student's graduation and publication are not subject to commercial terms. That is the premise, not a concession.
我们最近在做什么What we have been doing
这一栏按时间倒序，只记录真的发生过的事。Reverse chronological, and only things that actually happened.
-
开放课题上线：改为由中心开放现场、题目请高校老师与学生自己提出，主线收敛到 FDE（前向部署工程师）。首批以单实验室、单课题、单学生为起步单元。Open topics launched. We open the site; the questions come from supervisors and students. The through-line is the forward-deployed engineer. The first cohort starts with one lab, one topic, one student.
-
网站繁简双版上线：繁体与简体各自独立成页、可互相切换，方便内地与香港的老师、学生各看各的版本。Traditional and simplified editions. Each is its own page with a switch between them, so readers on either side get the version they read faster.
-
对外定位调整：切入点由行业分类改为深港两地使用频次最高的生活场景 —— 过关通勤、两地消费、就医配药、升学查询。Positioning updated. Our entry point moved from industry categories to the everyday situations people hit most often across both cities — crossing the border, paying on either side, filling a prescription, checking school rules.
一起把 AI 放进每天都会发生的事情里Let's put AI into what happens every day
合作、洽谈或咨询，欢迎随时来信，我们会尽快回复。For partnerships, projects or enquiries, write to us anytime — we'll reply promptly.
lxy@shengangai.com