20 Foundational GenAI Papers in 20 Weeks

I’m reading 20 foundational GenAI research papers in 20 weeks — and writing up every single one.

Not abstract skims. Full reviews: key ideas, enterprise implications, my honest take.

Why

I build AI systems for a living. But I kept noticing a gap between using the technology and understanding the research behind it.

That gap bites you when you need to make real architectural decisions — choosing between approaches, debugging something unexpected in production, evaluating whether a new technique is worth adopting. Intuition built on demos only gets you so far.

So I’m closing the gap. Publicly, because writing forces precision and because if you’re in the same position, you can follow along.

The reading list

I’ve organised the 20 papers into 8 phases. Each phase earns you the vocabulary for the next — read in order.

Phase 1 — Foundations You Can’t Skip (Weeks 1–3)

1. Attention Is All You Need — Vaswani et al., 2017
2. Scaling Laws for Neural Language Models — Kaplan et al., 2020
3. Training Compute-Optimal LLMs (Chinchilla) — Hoffmann et al., 2022

Phase 2 — The Modern Architecture Stack (Weeks 4–6)

4. LLaMA: Open and Efficient Foundation Language Models — Touvron et al., 2023
5. FlashAttention — Dao et al., 2022
6. GPT-4 Technical Report — OpenAI, 2023

Phase 3 — Alignment, Safety & Governance (Weeks 7–9)

7. Training Language Models to Follow Instructions (InstructGPT) — Ouyang et al., 2022
8. Constitutional AI: Harmlessness from AI Feedback — Bai et al., 2022
9. Direct Preference Optimization (DPO) — Rafailov et al., 2023

Phase 4 — Efficient Training & Adaptation (Weeks 10–11)

10. LoRA: Low-Rank Adaptation of Large Language Models — Hu et al., 2021
11. Textbooks Are All You Need — Gunasekar et al., 2023

Phase 5 — Retrieval, Grounding & Production Systems (Weeks 12–14)

12. Retrieval-Augmented Generation (RAG) — Lewis et al., 2020
13. Lost in the Middle: How Language Models Use Long Contexts — Liu et al., 2023
14. Efficient Memory Management for LLM Serving with PagedAttention (vLLM) — Kwon et al., 2023

Phase 6 — Reasoning, Agents & Frontier Capabilities (Weeks 15–17)

15. Chain-of-Thought Prompting Elicits Reasoning — Wei et al., 2022
16. ReAct: Reasoning and Acting — Yao et al., 2022
17. DeepSeek-R1 — Guo et al., 2025

Phase 7 — Efficiency at Scale (Week 18)

18. Mixtral of Experts — Jiang et al., Mistral AI, 2024

Phase 8 — Evaluation & Interpretability (Weeks 19–20)

19. HELM: Holistic Evaluation of Language Models — Liang et al., 2022
20. Scaling Monosemanticity — Templeton et al., Anthropic, 2024

The format

Each post follows the same structure:

• One-line summary — what the paper actually says
• Why it still matters in 2026 — context for engineers, not historians
• Key ideas, explained simply — the concepts without the notation wall
• What it means for enterprise AI — architectural implications, practical constraints
• My take — where I agree, where I’d push back, what I’d do differently

Written for engineers who ship, not academics who cite.

Week 1 is live

Attention Is All You Need →

All paper reviews are tagged paper-review.