Pick a shelf, not a syllabus
Same field, four very different reasons to be in it. Pull the spine that matches why you're here — the list underneath is scoped to that job.
Goal: build foundational intuition across ML, DL, NLP, CV, and RL — in the order the field actually discovered it.
Foundations & History
A Logical Calculus of the Ideas Immanent in Nervous Activity
The artificial neuron.
The Perceptron
First learning rule.
Perceptrons
Limits of linear separability.
Learning Representations by Back-Propagating Errors
Backpropagation formalized.
Optimization & Training
Efficient BackProp
Practical training tips.
Understanding the Difficulty of Training Deep Feedforward Neural Networks
Xavier initialization.
Adam: A Method for Stochastic Optimization
The default optimizer.
Batch Normalization
Enables deep network training.
CNNs & Computer Vision
Gradient-Based Learning Applied to Document Recognition
LeNet-5.
ImageNet Classification with Deep Convolutional Neural Networks
AlexNet.
Very Deep Convolutional Networks
VGG.
Deep Residual Learning for Image Recognition
ResNet.
An Image Is Worth 16x16 Words
ViT.
Sequence Models & Transformers
Long Short-Term Memory
The LSTM.
Learning Phrase Representations Using RNN Encoder–Decoder
GRU & Seq2Seq.
Neural Machine Translation by Jointly Learning to Align and Translate
Attention mechanism.
Attention Is All You Need
The Transformer.
BERT: Pre-training of Deep Bidirectional Transformers
Bidirectional pretraining.
Generative Models
Auto-Encoding Variational Bayes
The VAE.
Generative Adversarial Networks
GANs.
Wasserstein GAN
Stable GAN training.
Modern LLMs & Scaling
Improving Language Understanding by Generative Pre-Training
GPT.
Language Models are Few-Shot Learners
GPT-3.
Scaling Laws for Neural Language Models
Predictable scaling.
Theory & Robustness
Intriguing Properties of Neural Networks
Adversarial examples.
Understanding Deep Learning Requires Rethinking Generalization
Networks memorize random labels.
Goal: understand AI's strategic impact, ROI, competitive advantage, and organizational transformation.
Strategy & Competitive Advantage
AI Business Strategy: How Generative Models Are Reshaping Competitive Advantage
Framework for how GenAI transforms industry structure.
AI And Competitive Advantage In The Agentic Era
Six imperatives for CEOs, from proprietary data to decoupling growth from cost.
The Impact of Artificial Intelligence on Business Strategy
Six new sources of advantage; 75% use AI in operations, 1% reach maturity.
AI Adoption & Transformation
Artificial Intelligence and Business Strategy Towards Digital Transformation
Systematic review of 81 papers on AI, knowledge management, and decision-making.
The Business of Artificial Intelligence
Economic framing of AI as a prediction technology.
Artificial Intelligence for the Real World
The "cognitive technologies" taxonomy.
AI Economics & Labor
Automation and New Tasks
How AI changes labor markets and task allocation.
Prediction Machines
AI as a drop in the cost of prediction.
AI Ethics, Risk & Governance
The Malicious Use of Artificial Intelligence
Threat landscape for business risk planning.
Model Cards for Model Reporting
Transparency and responsible AI deployment.
Case Studies & Scaling
Machine Learning: The High Interest Credit Card of Technical Debt
Why ML systems fail in production.
Hidden Technical Debt in Machine Learning Systems
The full cost of ML beyond model accuracy.
Goal: find product-market fit in AI, build defensible products, and stay at the research frontier.
AI Product-Market Fit & Startup Strategy
What Is Product-Market Fit?
PMF as a spectrum, Nascent → Extreme; 42% of startups fail on "no market need."
Mastering Product-Market Fit: A Detailed Playbook for AI Founders
Start with a wedge, not a platform. Ship fast, integrate faster.
How to Achieve AI Product-Market Fit
Lessons from SageMaker, Bedrock, and Q; the PR/FAQ framework.
Frontier Research — Must-Read
Attention Is All You Need
If you read one architecture paper, read this.
Language Models are Few-Shot Learners
Emergence of in-context learning.
Scaling Laws for Neural Language Models
The compute-optimal scaling bible.
Training Language Models to Follow Instructions with Human Feedback
InstructGPT — RLHF for alignment.
Chain-of-Thought Prompting Elicits Reasoning in LLMs
Reasoning via prompting.
Large Language Models are Zero-Shot Reasoners
"Let's think step by step."
Reinforcement Learning from Human Feedback
The practical RLHF guide.
LLaMA: Open and Efficient Foundation Language Models
Open foundation models.
Constitutional AI: Harmlessness from AI Feedback
Self-supervised alignment.
Agents & Systems
Generative Agents: Interactive Simulacra of Human Behavior
Agent architectures.
A Survey on LLM Based Autonomous Agents
Comprehensive agent survey.
ReAct: Synergizing Reasoning and Acting in Language Models
Reasoning + action loops.
Efficiency & Defensibility
LoRA: Low-Rank Adaptation of Large Language Models
Parameter-efficient fine-tuning.
QLoRA: Efficient Finetuning of Quantized LLMs
Fine-tuning on consumer GPUs.
LLM-QAT: Data-Free Quantization Aware Training
Model compression.
Goal: ship production-grade ML systems, understand MLOps, and build scalable infrastructure.
MLOps & Production ML
Designing Machine Learning Systems
The canonical book on ML system design.
Rules of Machine Learning: Best Practices for ML Engineering
43 rules for production ML.
The ML Test Score: A Rubric for ML Production Readiness
Checklist for production readiness.
Hidden Technical Debt in Machine Learning Systems
The seminal paper on ML technical debt.
System Design & Infrastructure
TensorFlow: A System for Large-Scale Machine Learning
Large-scale distributed training.
GPipe: Easy Scaling with Micro-Batch Pipeline Parallelism
Pipeline parallelism for giant models.
PipeDream: Generalized Pipeline Parallelism for DNN Training
Another pipeline parallelism approach.
ZeRO: Memory Optimizations Toward Training Trillion Parameter Models
Memory-efficient training.
LLM Engineering & Serving
Attention Is All You Need
You cannot build modern NLP systems without this.
Fast Transformer Decoding: One Write-Head Is All You Need
Multi-Query Attention — critical for inference optimization.
Efficiently Scaling Transformer Inference
Memory bandwidth and throughput.
FlashAttention: Fast and Memory-Efficient Exact Attention
The standard for efficient attention.
Retrieval & RAG
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
The original RAG paper.
Dense Passage Retrieval for Open-Domain Question Answering
Dense retrieval.
Atlas: Few-Shot Learning with Retrieval Augmented Language Models
Retrieval + few-shot.
Monitoring, Drift & Reliability
Failing Loudly: Detecting Dataset Shift
Data drift detection.
Instrumentation, Observability & Monitoring of ML Models
ML observability.
Data Distribution Shifts and Monitoring
Production data shifts.
RLHF & Alignment Engineering
Training Language Models to Follow Instructions with Human Feedback
The InstructGPT recipe.
Direct Preference Optimization
Simpler than RLHF.
Quantization & Edge Deployment
GPTQ: Accurate Post-Training Quantization
4-bit quantization.
AWQ: Activation-Aware Weight Quantization
On-device LLMs.
SmoothQuant
INT8 inference.
The Sutskever Canon
One researcher, fifty papers, thirty years of the field's actual arc — from a 2005 proof about √2 to the GPT-4o system card. Organized by era, ranked by influence.
The Breakthrough Era
2012–2014| # | Paper | Year | Citations | Why it matters |
|---|
Sequence Modeling & NLP
2013–2015| # | Paper | Year | Citations | Why it matters |
|---|
Foundations & Adversarial Robustness
2013–2016| # | Paper | Year | Citations | Why it matters |
|---|
Reinforcement Learning & Games
2016–2019| # | Paper | Year | Citations | Why it matters |
|---|
Generative Models & Multimodal AI
2016–2021| # | Paper | Year | Citations | Why it matters |
|---|
The GPT Era & Large Language Models
2018–2024| # | Paper | Year | Citations | Why it matters |
|---|
Vision & Multimodal
2021–2023| # | Paper | Year | Citations | Why it matters |
|---|