Reading benchmark results carefully
Why task coverage, prompt format, sampling policy, and contamination checks matter more than a single aggregate score.
Independent reference
Clear notes on language model behavior
Short, practical summaries of evaluation methods, inference constraints, and the trade-offs behind model comparisons.
Current notes
Latency, throughput, and context
A compact framework for comparing first-token latency, sustained generation speed, memory use, and long-context behavior.
Repeatable model comparisons
How fixed inputs, recorded parameters, multiple runs, and explicit uncertainty make small evaluations more useful.