Reranks multilingual retrieval candidates to favour documents that are both semantically relevant and written in the same language as the query, using English-anchored relevance distillation and preference alignment; excels in language-coherence tests while remaining competitive on standard multilingual reranking benchmarks.
Transforms open-ended LLM optimization into self-verifiable reinforcement learning by turning tasks into proxy environments that produce deterministic, rule-based rewards. Proposes RLSVR and SpyRL — an information-asymmetric self-play scheme where agents vote to identify a preassigned spy, yielding verifiable rewards without human annotation. Demonstrated on summarization, creative writing and mathematical reasoning.
AI Dataset2026
Provides a public test split of multimodal financial GUI interaction examples for evaluating agents that convert instructions and screenshots into grounded UI actions. Includes step-level screenshots, dialogue history, an OpenAI-style computer_use tool schema, and JSON next-action references; training data available on request.
Measures how agent memory systems miss implicitly associated facts by introducing InMind, a 125-task benchmark with paired controls that separate stored-vs-retrieval vs knowledge gaps. Quantifies a large retrieval-interface blind spot and points to routing as the core open problem.
Co-evolves a solver skill and a rubric-generator skill for text-space LLM optimization under decoupled objectives to avoid rubric gaming without using gold rubrics. Solver updates use criterion-level feedback; generator updates use independent audits of requirement coverage and response discrimination.
Allocates token-level credit in rubric-conditioned GRPO by counterfactually replaying the same response under rubric and criteria-free prompts, using tokenwise log-likelihood contrasts to compute bounded, response-normalized weights that redistribute GRPO advantages without training an auxiliary scorer.
AI Dataset2026
Provides fixed-seed benchmark instances (prompts and agent-visible inputs) for ASI-Bench to run reproducible evaluations of LLM agents on scientific tasks. Includes four matched prompt levels (B1–B4) across 60 project-level tasks in 11 domains; excludes reference answers and private scorers; Apache-2.0 licensed.
Evaluates multimodal context learning across grounding, new information application, and knowledge acquisition using a 3,443-instance benchmark spanning science, finance, long documents, spatial reasoning, and web VQA; finds current multimodal models perform poorly (best score 0.2847) and analyzes failure modes.
AI Video Papers2026
Converts text prompts into physically consistent videos by synthesizing executable Blender programs as a process-level chain-of-thought and using a dual-engine pipeline (deterministic simulation draft + draft-conditioned video editor). Ships with a VideoCoCo-3K draft–instruction–target dataset and shows substantial gains in physical-consistency benchmarks.
Evaluates whether vision-language models can make actionable decisions for a physical body by decoupling decision-making from low-level motor execution. Introduces HumanCLAW-Bench with 1,218 long-horizon egocentric episodes across 41 indoor scenes and diagnoses a lack of embodied self-awareness in current VLMs.
AI Agent Papers2026
Designs and evaluates a foundation GUI agent that performs cross-platform GUI and CLI actions on real devices to complete long-horizon workflows. Emphasizes a unified action space, a large-scale real-device mobile runtime, an AutoResearch-style data flywheel, and online RL training across 10,000+ concurrent environments.
Estimates the visually attributable portion of a privileged teacher’s next-token corrections and reconstructs student-anchored training targets for multimodal on-policy distillation. Uses counterfactual teacher queries and a signed proxy to raise supported tokens and suppress refuted ones, improving fine-grained visual knowledge transfer across model scales.