TESTSLLM temperature explained: what actually changes from 0 to 1.5
Temperature reshapes the odds of every next token. What changes at 0, 0.7 and 1.5, why 0 still is not deterministic, and how to pick a setting per task.
PROMPTSStructured output without JSON mode: 7 patterns that hold up
Schemas in the prompt, sentinels, validate-and-repair loops and smaller output shapes. Patterns that make any model return parseable data.
PAPERSSparse attention explained: sliding windows, global tokens and block sparsity
Why full attention gets expensive, and the three main ideas that make long contexts affordable, with the trade-offs each one makes.
MODELSQuantization without tears: what you lose at 8, 4 and 3 bits
Smaller weights, faster inference, and a quality cost that depends far more on the task than on the bit count. A practical guide to choosing a level.
TESTSHow to run an 8B LLM on a laptop: hardware, quantization and tools
Memory maths, quantization choices, tool options and a quick way to predict speed. Everything you need to run a capable model locally.
BENCHMARKSModel graders are useful, if you calibrate them first
Position bias, verbosity bias and self-preference. How to measure a model grader against human labels, fix its rubric and keep it honest over time.
BENCHMARKSLost in the middle: why long-context models still forget, and how to test it
Needle tests are easy to pass. Real documents need several facts combined from different places. Here is how to measure long-context recall properly.
TESTSRAG chunking strategies compared: fixed, recursive, structural, semantic and parent-child
How you split documents decides what your retriever can find. Five strategies, their trade-offs, sensible defaults and a method to choose on your own data.
MODELSAI-assisted coding that survives code review: a practical workflow
Scope small, give real context, write the test first and review every diff like a stranger's. A workflow for shipping AI-written code with confidence.