v0.5.3

Improve resistance to reward hacking breaking the UI

v0.5.2

Fix issues that allowed agents to break the UI by reward hacking

v0.5.1

Fix minor bug in the install validation for knowledge base tasks (issue #119)

v0.5.0: Managed instance pool

• Much easier user experience via #111
• Users no longer need to create and setup personal developer instances to use WorkArena.
• We manage a pool of instances, and benchmarking load is balanced across these.

v0.4.4

Fixes issue #86 regarding dashboard and report tasks.

v0.4.3

Fix data filter bug in dashboard/report tasks

v0.4.2

• Disable password policies in the installer to allow tasks to run in new user profiles created on the fly

v0.4.1: WorkArena++

• New compositional tasks

v0.4.0: WorkArena++

• 600+ Complex, compositional tasks
• Bugfix in dashboard tasks

v0.3.1: dashboard tasks renaming

Changed dashboard task names to match those in the ICML paper

• ReportValueRetrievalTask -> SingleChartValueRetrievalTask
• ReportMinMaxRetrievalTask -> SingleChartMinMaxRetrievalTask
• DashboardValueRetrievalTask -> MultiChartValueRetrievalTask
• DashboardMinMaxRetrievalTask -> MultiChartMinMaxRetrievalTask