Weekly Best Of: OCNews
Aug 30 – Sep 06, 2026
2 top-scored articles
Aug 30 – Sep 06, 2026
2 top-scored articles
Generated: September 06, 2026 at 03:54 AM ET
The technical report presents UI-Venus-2, an open-source foundation multimodal GUI agent designed to automate digital tasks across mobile, web, and desktop environments via a unified closed-loop reasoning-action framework. To overcome deployment challenges, the project scales environment coverage to over 170 multilingual mobile apps and native desktop operating systems, utilizes a deep-research pipeline for function-grounded instruction generation, and employs trace- and sample-level evaluators with visual keypoints and multi-model voting for reliable reinforcement learning signals. UI-Venus-2 also incorporates safety-aware mechanisms to regulate consequential actions.
Keywords: UI-Venus-2, GUI agent, multimodal AI, task automation, foundation models
The paper introduces EULER, a multi-agent system for mathematical discovery that utilizes cross-domain transfers, or "bridges," as its search unit. For a given conjecture, EULER competes direct, adjacent-domain, and distant-domain routes, using six ordered stress tests to filter out invalid bridges prior to intensive search. Tested on 120 recent combinatorics conjectures, EULER generated 10 proofs, 3 refutations, and 45 scoped partial results. Ablation analyses demonstrated that the stress tests reduced incorrect conclusions from 9 to 3, with success driven by executable operation gains and valid returns rather than domain distance.
Keywords: Artificial Intelligence, Multi-Agent Systems, Mathematical Discovery, AI Capabilities, Automated Reasoning, EULER