Browserbase Hud Frontier Evals
- Source
- browserbase.com
- Author
- Browserbase
- Date
Context
A useful browser- test needs a stable answer even when the website around it changes. Browserbase and the company HUD describe, in a joint article, one of HUD's live-web tasks, which asks an agent to retrieve historical air-quality readings from the EPA's AirNow interactive map. Fixing the date in the past keeps the target answer constant, while navigation still tests the agent's ability to operate the site. The grader scores four values with separate weights, so finding the right monitoring site but missing the reading still earns partial credit. Recorded step-by-step traces let reviewers see what the agent did, which helps separate a model's navigation mistake from a website, infrastructure, or grading problem.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Checking sign-in…
Loading comments…