News

A team from Georgia Tech's Data Science at GT (DS@GT) ARC lab submitted a system for the CLEF 2026 CheckThat! Task 2 that ranks reasoning traces from large language models (LLMs) and predicts verdicts for numerical claim…

A new study evaluating large language models' ability to recognize unspoken beliefs through implicature recognition and cancellation finds that LLMs lag behind humans, especially in natural scenarios. Researchers created…

TabRank, a framework for training reasoning rerankers for tabular retrieval, improves Acc@10 by 30.5% on HybridQA, 15.2% on SQA, 52.9% on TabFact, and 13.1% on TATQA compared to the base model, according to a new arXiv p…

GPT-5.4 achieved the highest macro-average score of 62.55 on MyoCardBench, a new real-world benchmark for evaluating large language models in cardiovascular care, according to a study published on arXiv. The benchmark in…

Researchers at HKBU KnowComp introduce VisualPatchWorld (VPW), a method that represents world dynamics as code for planning, achieving 69.0% mean planning success and outperforming the strongest code baseline by 23.5 poi…

A new study from arXiv (arXiv:2607.25270v1) finds that activation steering in language models is substantially affected by the choice of source activations, with effective steering signals coming from execution-boundary …

Researchers propose CAST (Credit Assignment from Solver Teachers), a method that converts game solver state-value changes into turn-level signals to improve reinforcement learning for LLM agents in long-horizon games. In…

A new study from arXiv:2607.25335 shows that linguistic rules alone can serve as effective prompt compressors for large language models, reducing inference costs without requiring LM-based scoring at compression time. Th…

Researchers propose temporal-distance JEPA (TD-JEPA), a plan-aware representation learning method that mines directed temporal cost from reward-free trajectories to improve latent world model predictive control. Under lo…

A new reusable template, llm-wiki-memory-template, enables LLM agents to maintain persistent, interlinked wikis that preserve dead ends and walked-back claims, addressing the negative-result loss problem in collaborative…

Kernel Forge, an open-source agentic harness for LLM-based generation and optimization of CUDA kernels, achieves speedups of up to 2.83× on softmax in Gemma 4 E2B and 1.70× on group_norm in Stable Diffusion 3.5 Medium, o…

A new study introduces CaRE, a compute-aware evaluation framework for masked diffusion language models (MDLMs), revealing that current evaluation standards conflate algorithmic improvements with hidden choices of compute…

A new study applying the Petri automated auditing framework to Qwen3-30B-A3B finds that LLM scheming behaviors inversely scale with pretraining language coverage, with low-resource languages averaging 34.2% higher scores…

LinkedIn researchers present a unified semantic modeling framework powered by a small language model (SLM) to improve job understanding at scale, achieving significant performance gains in offline evaluations and online …

Researchers have developed ProcAgent, a fully on-device, agentic, vision-based procedural assistant for real-time adaptive guidance on a single NVIDIA Jetson AGX Orin. The system uses a propose-and-verify architecture co…

Researchers have developed NRFormer+, a spatio-temporal Transformer that achieves state-of-the-art accuracy in nationwide nuclear radiation forecasting, reducing sudden-change mean absolute error by up to 19.1% over the …

Researchers introduce Generative Topology Optimization (GenTO), a unified framework that trains a diffusion model on a large full-order topology dataset and iteratively steers the resulting topology distribution toward t…

A new hierarchical reinforcement learning framework called HOBA (Hierarchical On-Policy Bidding Agents) achieves a +3.6% increase in target cost in large-scale online advertising deployment, according to a paper on arXiv…

A new study testing GPT-5.4, Claude Sonnet 4.6, Gemini 2.5 Flash, and Qwen3-235B on 101 World Values Survey questions across 13 language-country slices found that prompt framing is a first-order determinant of cultural a…

Researchers propose SpecPrefetch, a parameter-efficient prefetching framework for offloaded sparse Mixture-of-Experts (MoE) inference that uses a shared lightweight adapter to predict next-layer expert candidates for asy…