Foundational Reading

Articles & Papers

34 must-read engineering posts, tutorials, and breakthroughs.

tutorial

The Illustrated Transformer

by Jay Alammar

The best visual explanation of how transformers work. Essential reading before building one.

Read article
tutorial

Attention Is All You Need (Annotated)

by Harvard NLP

The original transformer paper, annotated with working PyTorch code line by line.

Read article
tutorial

The Illustrated GPT-2

by Jay Alammar

Visual walkthrough of GPT-2's architecture — how autoregressive language models generate text.

Read article
guide

A Recipe for Training Neural Networks

by Andrej Karpathy

Practical tips for training neural networks. The most shared ML blog post ever.

Read article
insight

Yes You Should Understand Backprop

by Andrej Karpathy

Why understanding backpropagation matters. Don't just use autograd blindly.

Read article
insight

The Bitter Lesson

by Rich Sutton

The most important essay in AI. General methods + compute always win over human knowledge.

Read article
research

Scaling Laws for Neural Language Models

by Kaplan et al. (OpenAI)

How LLM performance scales with data, compute, and model size. Foundation of modern scaling.

Read article
guide

An Introduction to LLMs

by Lilian Weng

Comprehensive blog covering LLM training, RLHF, prompting, and emergent abilities.

Read article
tutorial

Neural Networks: Zero to Hero

by Andrej Karpathy

Full video course building neural networks from scratch in Python. The GOAT free course.

Read article
insight

What Is ChatGPT Doing?

by Stephen Wolfram

Deep, intuitive explanation of how LLMs work. Great for building mental models.

Read article
tutorial

The Illustrated Word2Vec

by Jay Alammar

Visual guide to word embeddings. How words become vectors that capture meaning.

Read article
tutorial

Understanding LSTM Networks

by Chris Olah

The clearest explanation of LSTMs ever written. Beautiful diagrams, perfect intuition.

Read article
research

Chinchilla Scaling Laws

by Hoffmann et al. (DeepMind)

Compute-optimal training — most LLMs are undertrained. Changed how labs scale models.

Read article
insight

The AI Revolution: The Road to Superintelligence

by Tim Urban (Wait But Why)

The most accessible introduction to where AI is heading. Changed how millions think about AI.

Read article
guide

How to Read a Paper

by S. Keshav

Three-pass approach to reading research papers efficiently. Essential skill for researchers.

Read article
guide

Practical Tips for Fine-Tuning LLMs

by Sebastian Raschka

Concrete advice on LoRA, QLoRA, data preparation, and evaluation for fine-tuning.

Read article
tutorial

Understanding and Coding Self-Attention from Scratch

by Sebastian Raschka

Step-by-step implementation of the self-attention mechanism in PyTorch. The clearest code-first attention tutorial.

Read article
tutorial

A Visual Guide to Attention Variants in Modern LLMs

by Sebastian Raschka

Comprehensive visual exploration of MHA, GQA, MLA, sparse attention, and hybrid architectures used in modern LLMs.

Read article
guide

Understanding Reasoning LLMs

by Sebastian Raschka

Four main approaches to building reasoning-enhanced models — inference-time scaling, RLVR, and beyond.

Read article
tutorial

Improving LoRA: Implementing DoRA from Scratch

by Sebastian Raschka

The LoRA successor — Weight-Decomposed Low-Rank Adaptation explained and implemented in code.

Read article
guide

Components of a Coding Agent

by Sebastian Raschka

How coding agents work — tools, memory, and repository context for LLM-based agents. (Apr 2026)

Read article
insight

The State of LLMs 2025

by Sebastian Raschka

Annual review covering DeepSeek R1, RLVR, benchmarks, open-weight progress, and 2026 predictions.

Read article
guide

Understanding Multimodal LLMs

by Sebastian Raschka

Techniques and latest models in vision-language integration — how LLMs learn to see.

Read article
tutorial

Gradient Descent Visualized

by Lilian Weng

Visual guide to optimization algorithms — SGD, Momentum, Adam, and beyond.

Read article
insight

The Unreasonable Effectiveness of RNNs

by Andrej Karpathy

Classic blog showing what RNNs can learn — Shakespeare, code, Wikipedia, math.

Read article
tutorial

LLM Evaluation: 4 Approaches from Scratch

by Sebastian Raschka

Multiple-choice benchmarks, verifiers, leaderboards, and LLM judges — all with from-scratch code examples. The complete evaluation toolkit.

Read article
guide

LLM Architecture Gallery

by Sebastian Raschka

Visual gallery of 40+ LLM architectures (Llama, DeepSeek, Qwen, Gemma) with clickable diagrams, fact sheets, and concept explainers for GQA, MLA, and more.

Read article
research

Defeating Nondeterminism in LLM Inference

by Thinking Machines Lab

Why LLMs give different outputs even at temperature=0 — and an open-source PyTorch library for 100% bitwise-identical outputs. Essential for production reliability.

Read article
insight

Context Engineering is the New Skill

by Phil Schmid

Why context engineering has replaced prompt engineering as the key skill — how to structure inputs for agents, not just chat.

Read article
guide

Getting the Most Out of Claude Code 2.0

by Sankalp (dejavucoder)

The most bookmarked guide (20K bookmarks) on working with AI coding agents. Practical workflows and best practices from extensive real-world usage.

Read article
guide

Perplexity's 42-Page Internal AI at Work Guide

by Perplexity AI

How a top AI company actually uses AI internally — email automation, meeting prep, research amplification. 21K bookmarks. Real playbook, not theory.

Read article
guide

300+ ML System Design Case Studies

by Engineer1999

Battle-tested ML system designs from 80+ companies (Netflix, Spotify, Uber, Stripe). The best resource for understanding how ML works in production.

Read article
tutorial

Building a Vector Database from First Principles

by Archie Sengupta

From-scratch implementation with linear scan, KD-tree, HNSW, and IVF indexes. Understand how vector search actually works under the hood.

Read article
research

SWE-CI: Why AI Coding Agents Fail at Long-Term Maintenance

by Sun Yat-sen University & Alibaba

First benchmark testing AI agents across 233-day maintenance windows. 75% of models break previously working code. Only Claude Opus stays above 50%.

Read article