The Beauty of Large AI Models (GeekTime column): Review

Who’s writing this

I’m Lin Zhou, a Chinese full-stack developer turned AI application builder. I know the domestic developer-education ecosystem well, and The Beauty of Large AI Models deserves a particularly time-sensitive review: in 2023 it was close to the only systematic Chinese-language column on LLM application development; in 2026, its frozen syllabus changes the verdict. I’m writing in English for one reason: to tell you plainly who this course is for — Chinese readers — and why.

What it is

A paid column on GeekTime (Geekbang), one of China’s mainstream developer-education platforms, by Xu Wenhao, founder of the AI chatbot startup bothub and author of another column on classic big-data papers. Thirty-four installments: text plus audio plus runnable code. Pricing is dual-track — the official page showed a promo around ¥68 against a ¥199 list price as of August 2026, promos rotate, and it’s included in GeekTime VIP, so check current pricing. The decisive status fact: the column finished in July 2023 and has never been updated since. It still sells normally, with about 34,700 learners shown on the page, but it’s a completed, frozen product.

The column is entirely in Chinese — text with Chinese audio narration, no English version. Non-readers can stop here.

The content’s actual quality

Three modules, all application-layer: LLM fundamentals and prompt engineering; hands-on OpenAI API work (Completion endpoints, building a chatbot) plus embeddings for classification, clustering, and summarization; then AI retrofits of search and recommendation systems, AI-assisted reading and development (auto-generating unit tests), letting AI orchestrate external systems; and finally speech recognition and synthesis with lip-synced digital humans, Stable Diffusion image generation, and a combined “chat + draw” assistant project. For 2023, that was full-stack application coverage. Every installment ships runnable code, and Xu has real startup battle experience — he isn’t a career lecturer. As an intro survey, the system holds together. I still stand by that.

The one hard problem

Currency. Frozen since July 2023, the examples sit on GPT-3.5-era APIs and ecosystem assumptions — 2026’s model capabilities, agent paradigms, and toolchains are absent. Chinese-platform cross-reviews criticize GeekTime’s LLM catalog broadly for demos stuck on old interfaces and frameworks; here that criticism lands precisely. Depth stops at the application layer, with nothing on training or fine-tuning. Zhihu sentiment runs “decent intro, but a hype-wave course with average polish,” which matches my own read. At the platform level: slow official Q&A, no assignment grading, a final test added in July 2023, and no stated certificate policy. Running the code needs OpenAI API access, which mainland learners must arrange themselves — account and network — and that friction is unchanged in 2026.

How to buy it rationally in 2026

Browse it inside a membership, or catch a promo and treat it as a panoramic map of application development — fine. Buy it expecting a frontier course — don’t. Free same-topic alternatives: Datawhale’s llm-universe (still updated, same API + prompting + RAG stack) and DeepLearning.AI’s short courses (English, current). For principles over API calls, Hung-yi Lee’s NTU lectures.

Bottom line

The intro arc is complete, the code is runnable, and the instructor is credible. All of that was true in 2023 and remains true. The date is the problem: “finished in 2023, never updated” means visibly aged LLM content. A low promo price can still make sense; frontier-course expectations cannot. For English-only readers, none of this applies — the course was never for you.