AIGC-Interview-Book: a Chinese question bank organised by job track

The repository is one folder per topic and nothing to compile

There is no build step here, because there is no software. AIGC-Interview-Book is a documentation repository under the GPL-3.0 licence whose root is a set of topic directories, each holding Markdown notes:

AI Agent基础/
AIGC图像创作&AI绘画基础/
大模型基础/
AI多模态基础/
深度学习基础/
模型部署基础/
热门AI学习核心课程与教程/
数据结构&算法基础/
编程基础：Python/
编程基础：C和C++/
大厂高频面试题（实时更新）/

Alongside them sit `docs/`, `imgs/` for images, `workspace/`, a `LICENSE` file and the `README.md` you are reading. A rendered version is published at wethinkin.github.io/AIGC-Interview-Book/, which is the faster way to browse if your editor handles Chinese filenames badly.

The last push to the repository was on 2026-11-02, and it has no GitHub releases, so there is no version to pin and no changelog to read.

Note that the primary language is reported as unknown, because a Markdown repository has no dominant code language. Every file name above is Chinese, and so is the text inside them.

Three job tracks decide which folders are worth your time

The README splits its coverage into three audience tracks, and the split is more useful than a flat table of contents.

The algorithm track, `AIGC算法岗方向`, names AIGC, LLM大模型, AI Agent, 具身智能, 传统深度学习, 自动驾驶, 计算机视觉, 自然语言处理, 机器学习, 强化学习, 大数据挖掘, 世界模型, 元宇宙 and AGI. The development track, `AIGC开发岗方向`, is a different list entirely: FDE, Python, Java, C/C++, Go, 嵌入式, 前端, 后端, 测试 and 运维. The application track, `AIGC应用岗方向`, covers 产品, 运营, 设计, 商务 and 市场.

A backend engineer looking for Agent system design will find the answer in `AI Agent基础`, not in the development track's language list, and the two lists overlap only by accident. Read the track that matches the job you are applying for before you read anything else.

Beyond interview technique, the same README also promises 简历 diagnosis notes, AI大厂内推 referrals, AI薪资爆料 salary information and AI公司指南 company notes, which is the part most likely to age fastest.

Fifteen numbered routes live under 热门AI学习核心课程与教程

This is a folder of numbered Markdown files, each one a study route rather than a reference chapter. The numbering runs from 01 to 15 and some file names carry a time budget in the title.

`01_2026_AI_Agent岗面试50问.md` covers Agent essence, tool protocols, Memory, evaluation and platform design. `03_大模型面试最高频100问.md` covers architecture, training, post-training, RAG, inference and evaluation. Route files then vary by role and by transition: `05_从传统CV转AIGC的最短路径.md`, `06_30天AI算法岗冲刺路线.md`, `07_60天大模型算法岗系统路线.md`, `08_AI_Agent工程岗面试路线.md`, `11_模型部署与推理优化路线.md`, `13_开发岗转AI应用工程师路线.md`, `14_校招AI算法岗求职路线.md` and `15_社招AI算法岗跳槽路线.md`.

The 30 day and 60 day labels are a commitment, not a measurement. Nothing in the repository reports how long anyone took to work through them, so treat the day counts as the author's suggested pace.

One folder promises live updates, and several have 精华版 twins

Only one entry in the catalogue claims to change continuously: `大厂高频面试题（实时更新）`, described as high-frequency interview questions from large companies, updated in real time.

Several topics exist twice, once as a full version and once as an abridged one, and the root makes that visible:

AI视频基础（精华版）/
AI多模态基础（精华版）/
深度学习基础（精华版）/
模型部署基础（精华版）/

精华版 means essence edition. The catalogue's 大模型基础 entry links to 大模型基础（精华版）, although the root listing shows only 大模型基础/. The pairing is a practical one: read the short version first, then decide whether you need the long one. No rule is given for which topics are updated and which are frozen.

There is also a mismatch worth knowing about. `具身智能基础/` sits in the repository root, and embodied intelligence appears in the algorithm track list, but no catalogue entry links that folder. If you came for robotics, expect to browse the directory yourself.

The platform's argument is that most AI skills do not last

The stated premise is a claim about the labour market rather than about technology. The README says AI technology iterates fast but that not all technologies have durable cross-cycle value, and that only knowledge which has been repeatedly tested by industry, academia, competitions, investors and application teams is admitted as core content.

This is worth reading as an editorial position rather than as a finding. It is the filter the project applies to its own curriculum. The README also states the platform can serve as a reference source for university research, teaching and competitions, which is the audience most likely to accept that filter.

The editors are named, and so is the paid community behind them

Responsibility is attributed rather than anonymous. Rocky Ding is the editor in chief, described as founder of the AIGCmagic community and author of long-form notes on AI Agent and on diffusion models including DDPM, DDIM, Score-Based, SDE, LDM, Classifier-Free Guidance and Rectified Flow. 猫先生 is listed as a deputy editor.

The same README also promotes an advanced service community built from the AIGCmagic accumulation. Its stated offerings are 答疑 for technical questions, 内推 for referrals, 简历 diagnosis, 模拟面试 mock interviews, 项目辅导 and 职业规划. Access is through a WeChat QR code, the 知识星球 app is suggested for reading and posting, and entering the VIP group requires adding the assistant account `Jarvis8866` with a note giving 社区昵称 + 城市 + 从事/研究方向 + 公司/学校.

None of that is required to read the notes. It does mean the repository's front page mixes curriculum with recruitment services, so skim past the community sections before judging the notes.

No code, no citations, no English, and a China-shaped market

The limits are structural rather than fixable by a pull request. First, there is no code: not one compiler target, not one script, and no tests, so nothing here can be executed or verified by running it. Second, the README says the content is distilled from the editors' work, research and competition experience, and from written and interview questions at large technology and AI companies, and its catalogue lists no paper, benchmark or dataset references.

Third, everything is in Chinese, including every folder name, so a reader who does not read Chinese gets the structure and nothing else. Fourth, the advice is shaped for one market. 校招 and 社招 rhythms, 大厂内推 referrals, AI薪资爆料 salary figures and AI公司指南 company notes are all specific to Chinese hiring, and nothing translates them.

For an engineer outside that market the useful parts survive the translation problem: the topic decomposition, the role routes and the question sets. The career sections do not.

Editorial conclusion

Use AIGC-Interview-Book if you are preparing for an AI algorithm or engineering interview in the Chinese market, you read Chinese, and a curated folder structure saves you from assembling your own question list. Do not treat it as a technical reference for a specific framework, model or paper: there is no code, no bibliography and no English edition. Before you plan a preparation schedule around it, open `大厂高频面试题（实时更新）` and one file under `热门AI学习核心课程与教程/` to judge the depth against what you already know, because the project describes its own value judgement as editorial.

Frequently asked questions

Official sources

1. Issues
2. License: GPL-3.0
3. Project website
4. README
5. WeThinkIn/AIGC-Interview-Book on GitHub