Part 2 on summarizing a 200,000-word handbook: validating K-means clusters with UMAP and silhouette scores, picking the chunk nearest each centroid, and where the final reduce step still drops topics.
Levenshtein and Soundex break the moment a query crosses script boundaries. Notes from a sanctions screening build on byte-level transformer embeddings, ANCE hard negative mining, and where CJK still causes trouble.
Tabular RL breaks the moment the state space grows past a toy grid. Approximate Solution Methods swap the lookup table for a weight vector and semi-gradient updates, so one state teaches you about every state like it. Includes where linear models still beat a neural net.
The WordPress 7.0 Beta is live for Business and Commerce plans, and the change worth having is a one-click Deactivate button on the fatal error screen, instead of an SFTP session or a recovery mode email that never arrives.
Twenty-three pipeline failures in six weeks, none of them caused by my code. This is how I replaced GPT-4 with a seeded Qwen2.5 on Ollama, plus the GitHub Actions caching and Pydantic validation details that made it stick.
A house-pricing model with zero training error valued a suburban shack at $2.4 million. This is the geometry behind lasso: centering the data, the L1 diamond, and why landing on a corner zeroes a coefficient.
Naive before-and-after averages said the London tube strikes lifted bike usage by 5.5%. A two-way fixed effects model with clustered errors put it at 3.95%. This is why the second number is the one to trust.
Academic causal inference stalls in applied work. Decision gravity is the fix: constructive decisions can live with an associative read, final ones cannot, and collider bias is what catches the people who guess wrong.
Prefill is compute-bound, decode is memory-bound, and running both on one GPU wastes most of it. Splitting them into separate pools, the KV-cache transfer tax that comes with it, and when it pays off.
A synthetic dataset can pass every standard check and still lose the correlations, rare events and privacy margin that your model depends on. Here are the three checks I run before shipping one, with the Python for two of them.