ML Fundamentals

Most extractions break at merge, not at edit time. The build passes. The tests pass. The function...

A free model scores 30% on your eval. The obvious conclusion: the model is weak. Swap it, and the...

Your evaluation set has an expiration date. Last month it helped you pick a model. This month it is...

If you've ever had to instrument a coating line, you know the hard part isn't the optics. It's...

Classic Machine Learning Through the Eyes of an SRE — Part 9: Isolation Forest The algorithm in one...

A List can be considered as a dynamic array. It is denoted by [ ]. The values inside a list are...

Azure Machine Learning ties experiment tracking, pipelines, and model serving to Azure-specific APIs...

Free model endpoints are not free of mystery. You get a response, but no dashboard. No latency...

mcp-tool-sanitizer v0.1.0: Making the MCP approval-view match the bytes the model gets A...

We shipped a model update last quarter that moved our aggregate pass rate from 81.2% to 83.6%....

TL;DR: Judge-human agreement is almost always reported as one number over a whole validation set....

The One-Line Summary: I set out to write six posts comparing ensemble and tuning methods, and every...

Choosing between a Decision Tree and a Random Forest is one of the first architectural decisions in...

The most expensive place to call a paid model API is your own laptop during debugging. Every retry,...

I spent two months building a free GEO (Generative Engine Optimization) audit tool with a CLI and an...

FlashMorph slashes the cost of designing hybrid attention models, needing only 20 M tokens and...

I ran the $0.14 model against the $0.44 model expecting a close fight. It lost 4-0. Then I looked at what it does to everything priced in between.

The One-Line Summary: Voting has no learned parameters, so it cannot overfit and it cannot lie to...

The mismatch inside standard on-policy distillation On-policy distillation trains a...

A Newtonian N-body engine built around a sixth-order Yoshida integrator, an independent Decimal...

In the current epoch of Artificial Intelligence, the industry remains singularly preoccupied with the...

Streaming changes everything. Or so I thought. Then I measured it. The first token is a...

I measured Go 1.27's experimental simd package against NumPy. They tie out of cache and lose inside it, and the reason is not the language.

I measured Go 1.27's experimental simd package against NumPy. They tie out of cache and lose inside it, and the reason is not the language.

Every baby tracker shows growth percentiles. "Your daughter is in the 72nd percentile for weight." It...

Route most requests to the cheapest model that passes your evals, and send a request to the expensive model only when a cheap, checkable signal says the...

Combining 13 technical indicators in an XGBoost model with Bayesian-optimized hyperparameters - a complete Python pipeline.

When we write: const result = add(10, 20); Enter fullscreen mode Exit fullscreen...

Context: A model's weights — the numbers it uses to reason — are normally stored at high precision,...

If you build features for Numerai Signals from fundamentals, the single most common way to silently...

Choosing the Right GPU for Your Model — A Sizing Method, Not a Guess OK,...

The short version, for anyone who does not benchmark models for a living Every few weeks a...

Predictive System Health Checks: What I Learned Testing ARIMA, SARIMA, and Prophet on Infrastructure...

The Quest Begins (The "Why") Honestly, I used to stare at a blank editor and feel like...

Don't just ask for the answer. Ask how it got there. A model will hand you a conclusion with total...

A practical guide to building reliable event-driven systems in Go using Apache Kafka. Learn how to...

You do not need a large budget to find out whether a freshly announced model fits your system. A...

Text filtering looks deceptively simple when the first version works on isolated examples. Give a...

Consider a small team that sees two model releases in the same week. One is DeepSeek-V4-Pro-0813,...

Phase 2 of 8: Enough ML to understand inference. Week 5 of 32. Phase 1 was about running models....

Late on a Tuesday, a C++ tooling team noticed their warning classifier was duplicating...

Keeping an AI model fast enough to use is turning out to be the expensive part of building...

GitHub Models is now retired I missed this news until today, when the GitHub Actions run for my simonw/research repository failed with this error message: GitHu

Clients ask me why we don't just build our own model. It's a fair question, and most of the time the honest answer is that building our own would be the slowest

Unleash Your AI: Serving Models Like a Pro with TensorFlow Serving So, you've poured your...

The problem of speech recognition in contemporary artificial intelligence systems can be described as...

After bombing a bitwise operations question, I built a free visual calculator that shows AND, OR, XOR, NOT, NAND, NOR, XNOR with binary alignment.

This is part five of a series about pointing an append-only audit log at things that count tokens....