Training Computer Use Models In The Real World With Microsoft
- Source
- browserbase.com
- Author
- Browserbase
- Date
Context
Browserbase’s of Microsoft’s Fara-7B browser-control model makes the testing procedure central to interpreting the result. Browserbase is itself Microsoft’s exclusive partner for training and evaluating these models, and its account describes 595 web tasks, a shared browser environment and human review of screenshots, action histories and final task states. It also allowed up to five retries. Checking a different model, UI-TARS 1.5-7B, Browserbase reports finding tasks that an automated judge accepted even though a human considered them incomplete. A success score therefore depends on what counts as completion and how many opportunities the receives; comparing browser agents requires matching those conditions.
- 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.
- 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.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…