Resources

Code-first explanations of LLMs from scratch. The best for building deep intuition.

Deep paper reads with live commentary. Strong research taste, goes beyond the abstract.

Implementation-focused — watches you build Transformers, VAEs, etc from scratch in PyTorch.

Fast 2-5 min overviews of new papers. Good for staying current, not for depth.

Beautiful visualizations of math/ML fundamentals. The neural network series is required viewing.

Best Chinese ML content. Paper reading sessions, course lectures, intuitive explanations.

In-depth interviews with AI startup founders and researchers — Kimi, DeepSeek, major tech AI leads. Also on Xiaoyuzhou podcast.

Science and tech explanations with great visualizations. Covers AI concepts accessibly for a broad audience.

Chinese re-upload of 3Blue1Brown's animation videos.

Systematic ML course series, covers fundamentals to advanced topics.

Xiaohongshu has a large volume of AI/ML study notes. Search: "大模型", "AI科研", "论文精读", "机器学习入门" to find creators.

Weekly AI newsletter by Andrew Ng. Balanced coverage of research + industry.

Deep technical dives. Best for understanding recent LLM research thoroughly.

Weekly policy + research mix. Nuanced takes on AI safety and governance.

Long-form technical articles. High quality, peer-reviewed content.

Top Chinese AI media. Fastest news coverage, industry updates.

Chinese AI news + analysis. Covers both research papers and product launches.

Chinese ML paper summaries and reading groups. Community-driven.

Sharp takes on AI industry trends, startup moves, and product launches from a Chinese tech insider perspective.

Deep technical posts on RL, diffusion, and attention. The best technical summaries of entire subfields.

NLP and transfer learning deep-dives. Essential reading for language model researchers.

Long-form AI research journalism. High-quality editorial coverage of the field.

Classic posts like 'The Unreasonable Effectiveness of RNNs'. Rare but gold posts on LLMs and neural nets.

ML systems and LLM applications. Practical, engineering-focused perspective.

Deep mathematical derivations in Chinese. The go-to for understanding the theory behind attention, diffusion, etc.

One of the best Chinese explanations of BERT and pre-trained language models.

Long-form interviews with top AI researchers. Better for inspiration than technical depth.

This Week in ML & AI — weekly research interviews covering the breadth of the field.

Technical ML podcast, researcher-first perspective. Challenging and rewarding.

Chinese AI news and trends. Good for keeping up with the Chinese AI ecosystem.

Tech startup founder interviews including AI content. Practical builder perspective.

Interactive ML articles with visualizations (attention, CNNs, etc.). Inactive since 2021 but all articles remain gold.

Visual math explanations for neural networks. Best companion for learning the fundamentals.

Interactive GPT-2 walkthrough in your browser. Visualize how tokens flow through the transformer.

Visualize BERT/GPT attention heads. Great for building intuition about what attention actually learns.

Official PyTorch learning path. The most reliable way to learn the framework from the ground up.

Free NLP/diffusion courses with code. Practical and up-to-date with current ecosystem.

Practical DL course with a top-down approach. Start coding immediately, understand theory later.

MIT course covering the tools CS programs never teach: shell, scripting, Git, Vim, tmux, debugging, profiling, and more. Essential for anyone who spends time at a terminal.

ML practitioner blog with great tutorials on training, experiment tracking, and LLM fine-tuning.

Understand LLM training from scratch — no black boxes. Through controlled experiments, deeply understand every design choice in LLMs.

Interactive visual walkthrough of the Transformer architecture — attention heads, positional encoding, encoder-decoder flow, built from scratch.

Eric Raymond's classic guide on how to ask technical questions effectively in open source communities. Required reading before posting on Stack Overflow or any mailing list.

Uses DP and greedy algorithm metaphors to argue that always picking the locally optimal option traps you in a local optimum. Recommends introducing random perturbations to escape it.