Natural Language Processing

Researchers at Tevatron released BrowseComp-Plus, a new benchmark designed to evaluate deep-research AI agents by isolating the effects of retrievers and large language models for fair and reproducible comparisons. The b…

Researchers have developed a system that automatically discovers an executable schema from raw multi-source data, using it as a shared contract for knowledge graph construction and query-time retrieval. The system improv…

Researchers have developed camroll-agent, a conversational AI assistant that accesses a user's personal camera roll to answer questions about their photos, from factual queries like identifying food eaten yesterday to op…

Researchers at arXiv propose a hybrid pre-training objective for language models that combines a Joint Embedding Predictive Architecture (JEPA) latent-space prediction loss with masked language modelling (MLM). Pre-train…

Researchers developed a variance-aware reward framework using Group Relative Policy Optimization (GRPO) to improve heart-focused medical question answering in large language models. The approach, which replaces tradition…

Researchers introduced generic triple-latent sequence models that maintain a running token state and compressed pair-memory pathway to capture higher-order token interactions without benchmark-specific parsing. The tripl…

Researchers at an undisclosed institution conducted a systematic study of parameter-efficient fine-tuning (PEFT) using Low-Rank Adaptation (LoRA) on the Qwen2.5-3B model for telecommunications customer support, generatin…

Researchers have developed a framework combining Shapley-value attributions with large language model rationales to explain automated rubric-based scoring of teaching quality. Testing on 6,000 classroom transcript segmen…

Researchers have developed the Multi-Granularity Reasoning Network (MGRN), a new framework for natural language inference that explicitly leverages hierarchical semantic features across multiple levels of language unders…

Researchers have developed LANTERN, a lightweight memory layer that recovers facts lost when large language models compress long conversations, achieving 78.3% recovery of verifiable facts without requiring any LLM calls…

Researchers introduced trajectory extrapolation error, a measure of how much a language model's hidden states deviate from a linear path during word processing, and found it independently predicts human reading times bey…

Researchers introduced a two-stage early failure alerting system for dialogs and LLM-agent trajectories that learns from sparse turn-level evidence rather than assigning terminal labels to every prefix. The attention-bas…

Researchers have introduced state commitment learning, a new training objective that teaches language models to distinguish temporary computation from persistent memory. The method, called Counterfactual Erasure RL (CERL…

Researchers have developed CVT-RL, a reinforcement learning algorithm that uses policy-conditioned counterfactual credit assignment to reduce unsupported evidence chains and shortcut actions in long-horizon language agen…

Researchers have developed BUMP, a self-supervised framework that trains large language models to generate personalized user profiles without requiring labeled data from downstream tasks. The system uses a bidirectional …

Researchers have developed a non-autoregressive punctuation restoration method for streaming automatic speech recognition (ASR) systems that uses a weighted lookahead scoring approach with bounded future context. The met…

Researchers have developed LoRi, a low-rank distillation framework that improves implicit reasoning in large language models by aligning teacher and student reasoning trajectories within a shared low-rank tensor subspace…

Researchers have developed PERSUASIONTRACE, a framework that tracks human belief changes across multiple turns in conversations with large language models, revealing that people cluster into two distinct groups of belief…

Researchers have developed ReasoningFlow, a framework that maps the non-linear discourse structures of large reasoning model (LRM) traces into directed acyclic graphs (DAGs) to improve evaluation and monitoring. Through …

A new study introduces Errorquake-10k, a benchmark that scores LLM responses on a continuous 0-4 severity scale, revealing that open-weight models with matched accuracy differ substantially in their error severity distri…