Definition
Plain language
The best result anyone has published so far on a given task.
As stated in the literature
Current best reported performance on a benchmark; claims require comparison against properly tuned baselines, not only weak ones.
Also called: state-of-the-art, SOTA
Why it matters: The label drives attention and funding, so it is easy to claim by comparing against a sloppily configured competitor rather than a genuinely strong one.
For example, if the best published score on a reading test is 88 percent, a new method claiming 89 percent is the new state of the art.
Heard on the show
“The previous open-model state of the art, a system called WebRL, was at thirty-eight point four.”Episode 008 — Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps