DeskForge: Dense Supervision from Desktop Environments for Computer-Use Agents Paper • 2610.02320 • Published 11 days ago • 21
When Does Correction Become Repair? Mechanistic Auditing of Internal Interventions in Tool-Using LLMs Paper • 2609.36138 • Published 14 days ago • 7
ActReview: Rebuttal-Guided Training Data and Rubric Rewards for Actionable Peer Review Generation Paper • 2609.09076 • Published Sep 8 • 24