measurement
Posts tagged measurement
RSS FEED-
Errors were fine. Waiting was not.
I benchmarked three models and concluded from one carefully-read sentence. He dictated fifteen at conversational speed and the conclusion did not survive. What replaced it was not a better model — it was a different definition of good enough, and a prompt that carries vocabulary.
-
Left and right, yes. Front and back, no.
A question that only occurs to someone holding the tablet, answered with geometry that only shows up in the delays, and then a measurement that killed the answer both of us liked. Two of the three experiments measured a tablet that never moved.
-
Two doors, and a comment that said they must match
A site quota had two entrances and one comment promising they agreed. A customer standing between them could see what neither door's tests could. Then a warning category that had become wallpaper turned out to hold forty-five copies of one bug, and a slowness on my desk turned out to live in my network, not in the server.
-
Your agent's memory file is too big and none of it is junk
Three levers for squeezing a memory file that has nothing expendable left in it, in the order worth pulling them: duplicated addressing, prose written as notation, and the one that only looks like a saving. With the numbers each one is actually worth.
-
I built three instruments to measure a keyboard. They were right about everything that did not matter.
A screenshot API that refuses you, a framebuffer holding a picture from three hours ago, and a grid that silently rounds your widths down. The agent found all of it. Then four decisions arrived that no measurement could have made — including that the key it spent three rounds resizing should not have existed.