Appearance
Curated Resource List
The AI hot-concepts space spawns hundreds of new projects, courses, and leaderboards every year — far more information than any one person can digest. This page curates, along six dimensions — official docs and APIs, learning courses, must-read books, high-quality blogs, open-source tools and frameworks, and communities and leaderboards — the resources genuinely worth your time, each with a one-line positioning and its official entry point. Every link was verified one by one (as of June 2025).
How to Use This List
Use this list alongside this site's Learning Paths: first build your conceptual skeleton here (start with What Are AI Hot Concepts and the Glossary), then treat the resources on this page as an "external supply station." Don't bite off more than you can chew — mastering one main line beats bookmarking a hundred links.
1. Official Documentation and APIs
The first step in building LLM applications is "getting good with APIs." Each major vendor styles its docs differently, but all of them ship the same trio: Quickstart, API reference, and best practices. Learning to read official documentation is more reliable than chasing any secondhand tutorial.
| Name | Type | One-line intro | Link |
|---|---|---|---|
| OpenAI Platform | Official docs + API | The authoritative docs for the models and APIs behind ChatGPT, covering the GPT family, multimodality, fine-tuning, function calling, and the Realtime API | platform.openai.com/docs |
| Anthropic Docs | Official docs + API | The official entry point for the Claude model family, and the first place features like Claude Code, prompt caching, and MCP support are documented | docs.claude.com |
| Google Gemini API | Official docs + API | Official developer documentation for the Gemini models (including the extra-long 1M context), with AI Studio for debugging in the browser | ai.google.dev |
| DeepSeek API | Official docs + API | Official API documentation for DeepSeek, the open-source model from China — low prices and self-hostable open weights make it a high-value engineering pick | api-docs.deepseek.com |
| Alibaba Cloud Model Studio (Tongyi) | Official docs + API | The official Chinese entry point for the Qwen model family, a one-stop platform for model services, RAG building, and agent orchestration | help.aliyun.com/zh/model-studio |
| Mistral AI Docs | Official docs + API | Europe's leading open-model vendor — concise docs and self-hostable models, a good fit for "open source first, API second" architectures | docs.mistral.ai |
How to Read Docs
Spend 30 minutes getting the Quickstart running, then hit the API Reference with specific questions, and finally adapt the official Cookbook/Examples as templates for your own code. Do not read the docs cover to cover. For a side-by-side comparison of each vendor's model lineup, pricing, and context windows, see Models & Leaderboards Quick Reference.
2. Learning Courses
The value of a course lies in its "systematicness": someone has already sequenced the material for you. Free, high-quality courses are exceptionally dense in this field — just pick one at each of the three tiers: intro → application → fundamentals.
| Name | Type | One-line intro | Link |
|---|---|---|---|
| Andrew Ng's ChatGPT Prompt Engineering for Developers | Short video course (~1.5 hours) | DeepLearning.AI's most classic free course — teaches the key prompt engineering techniques through code examples, the ideal first LLM application course | deeplearning.ai course page |
| Building Systems with the ChatGPT API | Short video course (~1 hour) | Andrew Ng's follow-up course with OpenAI on assembling single calls into usable conversational systems and multi-step chains | deeplearning.ai course page |
| Hugging Face NLP Course | Free text-and-image course | From fine-tuning BERT to training your own LLM — the official teaching track built entirely on the Transformers ecosystem, essential reading for open-source model beginners | huggingface.co/learn/nlp-course |
| fast.ai Practical Deep Learning for Coders | Free videos + book | Top-down teaching — "get the model running first, then take it apart" — with your own model trained within a few lessons; the hands-on learner's first choice | course.fast.ai |
| Mu Li's Dive into Deep Learning | Free book + videos | The definitive hands-on deep learning textbook where every formula ships with runnable code, complete with full video walkthroughs on Bilibili | zh.d2l.ai |
| Karpathy's Neural Networks: Zero to Hero | Free video series | Builds GPT from scratch entirely by hand (including the tokenizer and backpropagation) — the clearest explanation of how the inside of a large model actually computes | karpathy.ai/zero-to-hero.html |
How to Choose Courses
Zero background, want to build LLM apps fast: take Andrew Ng's prompt engineering course plus Building Systems, and you'll have a respectable demo within a day or two. Want to go deep on principles: work through the Hugging Face NLP Course to master fine-tuning, and Karpathy's videos to reach the bedrock. Chinese-speaking learners: go straight through D2L. No need to finish them all — pick one main line.
3. Must-Read Books
A book's value isn't in being "new" — it's in being complete and rigorously derived. LLM books go stale fast, so when choosing, prioritize ones with a free official channel or whose authors are the authorities in the field.
| Name | Type | One-line intro | Link |
|---|---|---|---|
| Speech and Language Processing (Jurafsky & Martin) | Textbook | The Bible of NLP — the third edition is free to read online, spanning everything from language models to the Transformer and LLM era | web.stanford.edu/~jurafsky/slp3 |
| Deep Learning (the "flower book," Goodfellow et al.) | Textbook | The foundational work of deep learning — its first three chapters bring your linear algebra and probability up to speed in one go, and it's free to read online | deeplearningbook.org |
| Build a Large Language Model (From Scratch) (Sebastian Raschka) | Hands-on book | Walks you through writing a GPT from zero — tokenization, attention, pretraining, fine-tuning, and the full alignment pipeline — with a fully open-source code repository | github.com/rasbt/LLMs-from-scratch |
| Understanding Deep Learning (Simon Prince) | Free textbook | A 2023 free deep learning education — beautifully illustrated and mathematically rigorous, adopted as course material by many universities, with a chapter on diffusion models | udlbook.github.io/udlbook |
| Deep Learning for Coders with fastai and PyTorch | Hands-on book | The companion book to the fast.ai course — everything from zero to image-generating models is runnable, and the full text is free on GitHub | github.com/fastai/fastbook |
Reading Advice
Look things up as needed rather than reading cover to cover: use chapters 1–5 of the "flower book" as a math dictionary, and SLP as an NLP dictionary. If you want to systematically follow one full training pipeline, Raschka's book is the best main line — it strings together the whole "pretrain → fine-tune → align" path and pairs perfectly with this site's Fine-Tune Your Own LLM.
4. High-Quality Blogs
Blogs are the sweet spot between "timeliness" and "depth": easier to read than papers, more current than courses. Nearly every post from the authors below circulates widely in the field.
| Name | Type | One-line intro | Link |
|---|---|---|---|
| The Bitter Lesson (Rich Sutton) | Classic essay | The father of reinforcement learning's famous 2019 thesis — "compute and data will always beat human priors" — essential reading for understanding how AI evolves | incompleteideas.net |
| Lil'Log (Lilian Weng) | Technical blog | Blog of an OpenAI researcher — her systematic surveys of LLMs, agents, diffusion models, and safety amount to a "free map of the literature" | lilianweng.github.io |
| Simon Willison's blog | Technical blog | An independent developer and observer who documents the latest moves in LLM tooling and ecosystem nearly every day, balancing engineering practice with industry commentary | simonwillison.net |
| Jay Alammar's The Illustrated Transformer | Visual blog | The most famous Transformer illustration on the internet — countless people's first lesson in Attention; the series also includes the Illustrated GPT-2 | jalammar.github.io/illustrated-transformer |
| Sebastian Raschka's blog | Technical blog | The author of that LLM book — weekly LLM commentary that keeps close tabs on the latest research in fine-tuning, alignment, and inference | magazine.sebastianraschka.com |
| Synced (Jiqizhixin) | Chinese news outlet | The Chinese AI media outlet with the highest-quality paper explainers in the Chinese-speaking world — a reliable relay when the English resources won't parse | jiqizhixin.com |
Reading Rhythm
Daily: 15 minutes skimming arXiv titles or Simon Willison. Weekly: one close read of a Lil'Log or Raschka survey. On demand: consult Jay Alammar's illustrated posts when needed. Blogs answer "what should I believe," papers answer "on what grounds" — the former is the entry point, the latter is the real capability.
5. Open-Source Tools and Frameworks
Building RAG, agents, and deployments ultimately means composing these tools. First get clear on what problem you're solving, then pick a framework — avoid using a framework for the framework's sake.
| Name | Type | One-line intro | Link |
|---|---|---|---|
| LangChain | Framework (LLM orchestration) | The most popular LLM application orchestration framework — chains together calls, prompts, memory, and tools; the largest ecosystem but with thick abstraction layers | python.langchain.com |
| LlamaIndex | Framework (data ingestion) | A framework focused on "getting data into the LLM" — its depth in document parsing, indexing, and retrieval beats LangChain, and it's the first choice for RAG | docs.llamaindex.ai |
| LangGraph | Framework (agent orchestration) | A graph-based agent framework from the LangChain team that manages multi-step tool calls as a state machine — the mainstream choice for agent development in 2025 | langchain-ai.github.io/langgraph |
| vLLM | Inference engine | The de facto standard for open-source LLM serving — PagedAttention multiplies throughput, and it's the default foundation for self-hosted models | github.com/vllm-project/vllm |
| Ollama | Local runtime tool | Run open-source models like Llama and Qwen locally with a single command — zero-friction for learning and prototyping | ollama.com |
| Hugging Face Transformers | Model library | The de facto standard library for open-source models — nearly every mainstream model has official weights and an interface here, and it's where research reproduction starts | huggingface.co/docs/transformers |
How to Pick a Framework
Getting started: run models locally with Ollama and write a few dozen lines of plain Python — don't rush into a framework. Building a RAG app: start with LlamaIndex; for the full walkthrough see Build a RAG App from Scratch. Building a multi-tool agent: LangGraph. Shipping to production: vLLM; for the full recipe see Deployment and Inference Optimization in Practice. Understand the principles with bare code first, then let frameworks speed you up.
6. Communities and Leaderboards
Judging "which model is stronger, which direction is worth chasing" is a job for leaderboards and communities, not social media feeds. The four entry points below cover four layers: papers, leaderboards, code, and discussion.
| Name | Type | One-line intro | Link |
|---|---|---|---|
| Hugging Face | Model community | The "GitHub" of open-source models and datasets — model cards, downloads, and hosted inference in one place, and the first home of frontier model releases | huggingface.co |
| LMArena (formerly Chatbot Arena) | Model leaderboard | A "blind-test voting" leaderboard maintained by Stanford, UC Berkeley, and others — the most trustworthy model ranking based on human preference | lmarena.ai |
| arXiv | Paper repository | Where all AI papers appear first — subscribe to the daily listings for cs.LG, cs.CL, and cs.CV | arxiv.org |
| Papers with Code | Paper index | The authoritative index for "does this paper have an open-source implementation, and where does it rank on which benchmarks" — check it before reading any paper | paperswithcode.com |
| GitHub Trending | Community | The hotspot chart for open-source projects — a daily glance is the fastest way to sense which way the tooling ecosystem's tide is flowing | github.com/trending |
How to Read Leaderboards
For head-to-head model comparisons, trust LMArena's human blind tests and this site's own Models & Leaderboards Quick Reference. But keep three things in mind: first, leaderboard scores are "statistical averages" and say nothing about your specific task; second, open-source models shuffle the rankings extremely fast, so never treat any single ranking as a permanent verdict; third, evaluation on your own task is the final judge — see Build an LLM Evaluation Suite for how.
How to Use This Page: Recommended Combos by Role
Different roles have different goals, so the resource mix should differ too. The table below lays out three "main-line combos":
| Role | Main-line combo | Suggested pace | Companion pages on this site |
|---|---|---|---|
| Beginner (new to the field, wants to understand what LLMs are and can do) | Andrew Ng's prompt engineering course → Hugging Face NLP Course → get an open-source model running locally with Ollama | 2–4 weeks; hands-on first, theory later | What Are AI Hot Concepts + Learning Paths + Glossary |
| Engineer (building RAG/agents/deployments, solving business problems) | Official API docs (pick one) → build apps with LlamaIndex/LangGraph → deploy with vLLM | Project-based cycles; docs first, blogs second | Build a RAG App from Scratch + Deployment and Inference Optimization in Practice + Common Pitfalls |
| Researcher (reading papers, reproducing results, tracking the frontier) | Karpathy's fundamentals series → Raschka's book cover to cover → regular close reads of arXiv + Lil'Log → reproduce with HF Transformers | A fixed 3–5 hours per week on papers | Paper Deep Dives + Models & Leaderboards Quick Reference + Frontier Progress |
One Last Piece of Advice
The most overlooked resource of all is the official documentation itself. Secondhand tutorials go stale, while docs keep updating with releases; whenever "the tutorial says one thing but the error says another," go back to the docs, check the changelog, read the source — that's what separates engineers and researchers from everyone else. When you need datasets for experiments, this site also maintains a Datasets & Tools Archive.
Further Reading
- Learning Paths: Three Routes — the master study schedule for the resources on this page
- Models & Leaderboards Quick Reference — a one-page comparison of models and APIs
- Datasets & Tools Archive — datasets for practice and experiments
- Glossary — look up unfamiliar terms you meet in these resources
- Build a RAG App from Scratch — the first project that turns resources into work
- Build an LLM Evaluation Suite — learn to evaluate models before you lean on these resources
- Paper Deep Dives: Start Here — the bridge from resources to the original papers
References
- DeepLearning.AI short course catalog — deeplearning.ai/short-courses
- Hugging Face learning hub — huggingface.co/learn
- Simon Prince's Understanding Deep Learning official page — udlbook.github.io
- Sebastian Raschka's Build a Large Language Model official repository — github.com/rasbt/LLMs-from-scratch
- LMArena official blog (leaderboard methodology) — lmarena.ai/blog
- arXiv official site — arxiv.org
- Ollama official model library — ollama.com/library