Why HCI Matters for AI Builders: Lessons from a CS6750 Review
Eugene Yan’s review of OMSCS CS6750 argues that user-centered design is a critical, often missing skill for machine learning engineers.
AI & ML
Frontier models, research results, agents, and developer tooling.
Why Writing Is a Core Skill for AI Practitioners
Eugene Yan argues that writing is not a final step but part of a continuous learning cycle essential for clarity of thought.
Automating ML Experimentation with Jupyter, Papermill, and MLflow
A practical guide to streamlining machine learning experiments by automating notebook execution and tracking results.
Data Scientist vs Data Engineer vs ML Engineer: Panel Clarifies Roles
A Google-hosted panel at a DataScience SG meetup tackles the often-blurred boundaries between three key data roles.
Zettelkasten Note-Taking Beats Regular Notes for Learning
A Zettelkasten system helps people learn and remember more by making explicit connections between notes, according to Eugene Yan.
Beyond Matrix Factorization: Graphs and NLP for Recommender Systems
A meetup talk compares simple matrix factorization baselines with newer graph- and NLP-based recommender approaches.
Obsidian Copilot: RAG Assistant for Writing and Reflection
A practical walkthrough shows how retrieval-augmented generation can turn a personal knowledge base into a writing and reflection copilot.
Neural Algorithmic Reasoning: Teaching Networks to Compute
A new article explores how neural networks can learn classical algorithms such as sorting and shortest-path finding.
How Attention Works in Transformers: QKV, Heads, and Layers
Eugene Yan offers a practical, intuition-first guide to the core mechanisms that let transformers focus on what matters.
Four Building Blocks for Production LLM Systems
Eugene Yan's AI Engineer keynote frames LLM system development around evals, retrieval-augmented generation, guardrails, and feedback collection.
Guest Post Shares 5 Lessons from a Year of Writing Online
A guest post on Eugene Yan's blog reflects on public writing, but the source does not include the lessons and is unrelated to AI/ML.
Match LLM Patterns to Problems: External vs Internal, Data vs Not
A practical framework for choosing LLM approaches based on whether a problem sits outside or inside the model and whether it is data-driven.
Out-of-Domain Finetuning Cuts Labeled Data for Hallucination Detection
A method for bootstrapping hallucination detectors with open-source data could reduce the need for task-specific labeled examples.
Fish Counting Shows Limits of AI in Sociotechnical Systems
A look at salmon counting reveals why AI deployment in complex fields depends on human oversight and institutional context.
What Makes Push Notifications Useful — and When to Stop Sending
A practical breakdown of how to keep push notifications helpful while limiting the annoyance that comes from over-sending.
Seven Recurring Patterns for Building LLM-Based Systems
A field guide to the common technical and product patterns that keep showing up in production LLM applications.
Evaluating Abstractive Summaries: Metrics and Hallucination Detection
A practical guide to assessing summary quality beyond ROUGE, covering reference, context, preference metrics, and self-consistency checks.
Counterfactual Evaluation: A Better Way to Test Recommenders
A new framework treats recommendation evaluation as an interventional problem, using inverse propensity scoring to correct for bias in historical data.