I test LLMs for practical problem-solving—extraction, comparison, verification—not for chasing hype. These are field notes from real benchmark runs, not regurgitated press releases.
Expect concise updates ('new version works, but mind the latency') and thorough analysis when the topic demands it.