02 — Newsletters & Blogs¶

The gap between what blogs claim is happening in ML and what’s actually happening in ML is measurable. Newsletters distill this, but most of them are optimized for click-through rates, not understanding. The ones listed here are chosen because practitioners — not marketing teams — wrote them, and because they hold up when you cross-check against the actual papers and codebases.

Subscription strategy: start with 3 newsletters maximum. Drowning in ML content is not the same as learning ML.

Newsletters — Ranked by Signal Quality¶

🟠 Tier 2: Reliable — Worth Subscribing¶

The Batch — Andrew Ng / deeplearning.ai

URL: deeplearning.ai/the-batch
Frequency: Weekly
Cost: Free
Focus: Research explained for practitioners. Industry applications.
Signal rating: ⭐⭐⭐⭐
Honest assessment: High production quality, reliable framing, but rarely pushes into technical depth. Ng writes well but safely. Good for staying oriented without going deep. His “Letters” section has genuine wisdom on navigating the field.
Best for: All 13 months as a weekly pulse-check. Not for depth.

Lilian Weng’s Blog — Lilian Weng (OpenAI)

URL: lilianweng.github.io
Frequency: Irregular (~4-8 posts/year)
Cost: Free
Focus: Deep technical dives: attention mechanisms, RL, diffusion models, agents
Signal rating: ⭐⭐⭐⭐⭐ (when she posts)
Honest assessment: These are arguably the best technical ML blog posts in existence. The “Attention? Attention!” post alone is worth more than most courses on transformers. Not a newsletter — bookmark and check periodically. She posts irregularly but every post is dense gold.
Must-reads: “Attention? Attention!”, “What is ChatGPT doing and why does it work?”, “Prompt Engineering”, “LLM-powered Autonomous Agents”

Chip Huyen’s Blog / Newsletter

URL: huyenchip.com
Frequency: Irregular (several posts/year)
Cost: Free
Focus: MLOps, ML systems design, career, production ML
Signal rating: ⭐⭐⭐⭐
Honest assessment: Author of “Designing ML Systems” (O’Reilly). Her blog posts are research-grade writing on production problems. “MLOps: What It Is, Why It Matters” and “A Friendly Introduction to Machine Learning Compilers” are required reading for applied engineers.
Best for: Months 7-13 (MLOps phase).

Eugene Yan’s Blog

URL: eugeneyan.com
Frequency: Bi-weekly newsletter + blog posts
Cost: Free
Focus: Applied ML at scale (Amazon), RecSys, production patterns, career
Signal rating: ⭐⭐⭐⭐
Honest assessment: Eugene ships things that serve millions of users and writes about the actual problems. “Patterns for Building LLM-based Systems and Products” is one of the best practical architecture references available.
Best for: Months 8-13. Production + systems mindset.

🟡 Tier 3: Useful for Scanning (Low Depth, High Volume)¶

TLDR AI

URL: tldr.tech/ai
Frequency: Daily
Cost: Free
Subscribers: 1.1M+
Signal rating: ⭐⭐⭐ (breadth, not depth)
Honest assessment: 5-minute daily scan. Good for not missing a major paper or product release. Zero technical depth. Think of it as a headline service, not a learning tool.

The Rundown AI

Frequency: Daily
Cost: Free
Subscribers: 2M+
Signal rating: ⭐⭐ (very broad)
Honest assessment: Good for business/application layer. Very surface-level. The ML engineer equivalent of reading TechCrunch.

AlphaSignal

URL: alphasignal.ai
Frequency: Weekly
Cost: Free
Focus: Trending repos, papers, tools
Signal rating: ⭐⭐⭐
Honest assessment: Good for tracking trending GitHub repos and “what’s hot this week.” Lacks depth but useful as a discovery tool.

Blogs — Not Newsletters, But Required Reading¶

These are not subscription newsletters — they’re archives that you mine deliberately.

Jay Alammar’s Blog¶

URL: jalammar.github.io
Status: Posts slowed significantly after 2022. Existing articles are still canonical.
Signal rating: ⭐⭐⭐⭐⭐ (archive)
Must-reads: “The Illustrated Transformer”, “The Illustrated BERT”, “Visualizing A Neural Machine Translation Model”, “The Illustrated GPT-2”
Honest assessment: Jay’s visual explanations are the single best introductory resource for transformer internals. If you don’t understand attention after reading “The Illustrated Transformer,” re-read it. It is not outdated for understanding mechanisms.
Best for: Months 3-6 (transformer fundamentals).

Sebastian Ruder’s Blog¶

URL: ruder.io
Status: Active. NLP focus.
Must-reads: “An Overview of Multi-Task Learning in Deep Neural Networks”, “The State of Transfer Learning in NLP”
Best for: NLP specialization track.

Andrej Karpathy’s Blog¶

URL: karpathy.github.io
Status: Rare posts, but each is a classic
Must-reads: “The Unreasonable Effectiveness of Recurrent Neural Networks”, “Hacker’s Guide to Neural Networks”, “Recipe for Training Neural Networks”
Honest assessment: “Recipe for Training Neural Networks” (2019) is still the most accurate checklist for debugging training issues in 2025. Save it.
Best for: Months 2-8. Keep re-reading as your understanding deepens.

⚠️ DISTILL.PUB — STATUS: INDEFINITELY INACTIVE¶

⛔ DISTILL.PUB IS ON INDEFINITE HIATUS.

The journal announced hiatus in September 2021. As of July 2026, no new articles have been published and no resumption has been announced. The archive at distill.pub remains accessible and valuable — but it is a static archive, not an active publication.

Do not add Distill.pub to a “follow this newsletter” list. It is not a newsletter anymore.

Do read the archive. Articles like “Attention and Augmented Recurrent Neural Networks,” “Feature Visualization,” “Circuits,” and “An Introduction to Bayesian Methods” are among the best technical ML writing produced between 2016-2021. They remain accurate and worth careful study.

Distill archive must-reads:

“Attention and Augmented Recurrent Neural Networks” (2016)
“Feature Visualization” (2017)
“The Building Blocks of Interpretability” (2018)
“Zoom In: An Introduction to Circuits” (2020)
“A Mathematical Framework for Transformer Circuits” (2021)

Newsletter Strategy for the 13-Month Plan¶

Phase	Months	Recommended
Foundations	1-3	Import AI (weekly), The Batch (weekly)
Deep Learning	4-6	+ Ahead of AI (read every issue)
Systems & Production	7-9	+ Latent Space, + Chip Huyen blog
Specialization	10-13	Ahead of AI + Eugene Yan + domain-specific

What most people get wrong: They subscribe to 10 newsletters and read none of them properly. Pick 2-3, read every issue, and actually follow the links to the papers mentioned. A newsletter you scan for 90 seconds teaches you nothing. One you spend 45 minutes dissecting teaches you the shape of the field.

Return to README.md · Previous: 01_online_communities.md · Next: 03_india_specific_communities.md