AI Research

EffiRAG, a graph-based retrieval-augmented generation system, produced the preferred answer on 93 of 120 open-ended questions in the UltraDomain benchmark while cutting total system cost by 57 percent, from USD 0.952 to …

Researchers introduced ContrAgent, a contract-based framework that uses assume-guarantee contracts in linear temporal logic over finite traces (LTLf) to supervise LLM agents, compiling each contract into a deterministic …

Researchers proposed WFM, a Wiki Foundation Model for scalable, agent-native knowledge representation and retrieval, detailed in arXiv paper 2609.18182v1. WFM formalizes a Wiki Graph schema coupling dense document contex…

Researchers introduced REPAIR, a self-evolving data augmentation framework that iteratively synthesizes training data to resolve long-tail concept confusion in scientific dense retrievers, according to an arXiv paper (ar…

Researchers introduced AutoTuneBench, a benchmark and measurement protocol that makes trust architectural for LLM agents auto-tuning GPU kernels and serving engines, after characterizing four failure modes from a four-da…

Researchers introduced AeroWeaver, an embodied-agent harness that weaves individual UAV skills into coordinated mission-level behavior, according to arXiv paper 2609.18520v1. AeroWeaver connects large language model sema…

Researchers released BENCHCOMPASS, a payment-domain benchmark for large language models built from typed evidence packs with expert-reviewed admission, according to the arXiv paper 2609.18270v1. Across 16 model variants,…

Researchers introduced LGM, a neuro-symbolic framework that disentangles long-term memory in a continuous latent space for personalized agents, according to an arXiv paper (arXiv:2609.18461v1). LGM uses a sparse autoenco…

A new arXiv paper (2609.18471v1) identifies a failure mode it calls Onset Refusal Collapse (ORC), in which the refusal-related signal of Large Reasoning Models drops sharply at the first generated token under harmful que…

Researchers proposed MAGER, a multi-agent genetic evolution framework that automatically discovers meta-paths optimized for LLM reasoning in fake news detection, according to the arXiv paper 2609.18597v1. MAGER compresse…

A new arXiv paper (2609.18515v1) introduces "cunning questions" — prompts with misleading premises, atypical reasoning, or subtle inconsistencies — as training data that improved large language model robustness to out-of…

A new arXiv paper, DyMT-ESB, introduces a controlled evaluation protocol for social bias in multi-turn user-LLM interactions that generates follow-up user queries from evolving dialogue history rather than relying on pre…

A study of on-device tool routing over 600 Korean and English requests and a catalog of 70 local actions found that abstention, not tool selection, is where a neural component is required, according to the arXiv paper 26…

A new benchmark called ReFigBench, built on 1,000 real overview figures retrieved from arXiv papers, evaluates multimodal coding agents on reconstructing scientific figures as editable PowerPoint slides. Testing ten conf…

Applying reinforcement learning with verifiable rewards to the GLM-4-Voice speech model raised free-form accuracy on the GSM8K math benchmark to 74.8%, a new state of the art for speech-native models, according to an arX…

Researchers proposed STRETCH (Self-Taught Reasoning Evolution via Targeted CHallenge), a unified framework that uses a dynamic Stretch Zone mechanism to align question difficulty with an LLM's evolving solving capability…

Researchers introduced PersonaPath, a benchmark pairing 2,000 fine-grained learner personas with a hierarchical knowledge graph of 347 textbooks, 1,751 units, and 4,092 concepts across 77 subjects, to evaluate Knowledge-…

Researchers proposed EviGen, a three-layer framework for verifiable clinical rationale generation that uses a patient-conditioned retriever with learnable queries, an LLM generator, and a process-supervised verifier, acc…

A new paper titled "EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents" addresses the limitation that large language model trading agents rely on static hand-written tool-use policies f…

Researchers introduced VC-Attention, a method combining value smoothing and softmax casting to enable low-bit attention for Diffusion Transformers used in video generation. The work targets the dominant deployment cost o…