Glossary · Term

WebRL

← all terms

Definition

Plain language

An earlier open-model system for training AI web-navigation agents, used as a comparison point.

As stated in the literature

A reinforcement-learning recipe for web agents that set the prior open-model state of the art on WebArena-Lite, surpassed by the milestone-reward MiRA approach.

Why it matters: Having a strong open prior method gives the field a clear yardstick for judging whether newer web-agent training approaches are actually better.

For example, when researchers report how well a new web-agent method clicks through online tasks, they may line up its scores against this earlier system to show the improvement.

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

Mentioned in 1 episode

  1. 008
    Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps

Related terms