Natural Language Processing

Language models, text understanding, machine translation, and speech

Includes:Computation and Language(cs.CL)

Trending in Natural Language Processing

Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agents

Large language model (LLM) agents face fundamental limitations in long-horizon reasoning due to finite context windows, making effective memory management critical. Existing methods typically handle long-term memory (LTM) and short-term memory (STM) as separate components, relying on heuristics or auxiliary controllers, which limits adaptability and end-to-end optimization. In this paper, we propose Agentic Memory (AgeMem), a unified framework that integrates LTM and STM management directly into the agent's policy. AgeMem exposes memory operations as tool-based actions, enabling the LLM agent to autonomously decide what and when to store, retrieve, update, summarize, or discard information. To train such unified behaviors, we propose a three-stage progressive reinforcement learning strategy and design a step-wise GRPO to address sparse and discontinuous rewards induced by memory operations. Experiments on five long-horizon benchmarks demonstrate that AgeMem consistently outperforms strong memory-augmented baselines across multiple LLM backbones, achieving improved task performance, higher-quality long-term memory, and more efficient context usage.

2601.01885

Jan 2026Computation and Language

MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory

The hallmark of human intelligence is the ability to master new skills through Constructive Episodic Simulation-retrieving past experiences to synthesize solutions for novel tasks. While Large Language Models possess strong reasoning capabilities, they struggle to emulate this self-evolution: fine-tuning is computationally expensive and prone to catastrophic forgetting, while existing memory-based methods rely on passive semantic matching that often retrieves noise. To address these challenges, we propose MemRL, a framework that enables agents to self-evolve via non-parametric reinforcement learning on episodic memory. MemRL explicitly separates the stable reasoning of a frozen LLM from the plastic, evolving memory. Unlike traditional methods, MemRL employs a Two-Phase Retrieval mechanism that filters candidates by semantic relevance and then selects them based on learned Q-values (utility). These utilities are continuously refined via environmental feedback in an trial-and-error manner, allowing the agent to distinguish high-value strategies from similar noise. Extensive experiments on HLE, BigCodeBench, ALFWorld, and Lifelong Agent Bench demonstrate that MemRL significantly outperforms state-of-the-art baselines. Our analysis experiments confirm that MemRL effectively reconciles the stability-plasticity dilemma, enabling continuous runtime improvement without weight updates.

2601.03192

Jan 2026Computation and Language

Fast-weight Product Key Memory

Sequence modeling layers in modern language models typically face a trade-off between storage capacity and computational efficiency. While Softmax attention offers unbounded storage at prohibitive quadratic costs, linear variants provide efficiency but suffer from limited, fixed-size storage. We propose Fast-weight Product Key Memory (FwPKM), a novel architecture that resolves this tension by transforming the sparse Product Key Memory (PKM) from a static module into a dynamic, "fast-weight" episodic memory. Unlike PKM, FwPKM updates its parameters dynamically at both training and inference time via local chunk-level gradient descent, allowing the model to rapidly memorize and retrieve new key-value pairs from input sequences. Experiments reveal that FwPKM functions as an effective episodic memory that complements the semantic memory of standard modules, yielding significant perplexity reductions on long-context datasets. Notably, in Needle in a Haystack evaluations, FwPKM generalizes to 128K-token contexts despite being trained on only 4K-token sequences.

2601.00671

Jan 2026Computation and Language

mHC: Manifold-Constrained Hyper-Connections

Recently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and diversifying connectivity patterns. While yielding substantial performance gains, this diversification fundamentally compromises the identity mapping property intrinsic to the residual connection, which causes severe training instability and restricted scalability, and additionally incurs notable memory access overhead. To address these challenges, we propose Manifold-Constrained Hyper-Connections (mHC), a general framework that projects the residual connection space of HC onto a specific manifold to restore the identity mapping property, while incorporating rigorous infrastructure optimization to ensure efficiency. Empirical experiments demonstrate that mHC is effective for training at scale, offering tangible performance improvements and superior scalability. We anticipate that mHC, as a flexible and practical extension of HC, will contribute to a deeper understanding of topological architecture design and suggest promising directions for the evolution of foundational models.

2512.248801

Dec 2025Computation and Language

K-EXAONE Technical Report

This technical report presents K-EXAONE, a large-scale multilingual language model developed by LG AI Research. K-EXAONE is built on a Mixture-of-Experts architecture with 236B total parameters, activating 23B parameters during inference. It supports a 256K-token context window and covers six languages: Korean, English, Spanish, German, Japanese, and Vietnamese. We evaluate K-EXAONE on a comprehensive benchmark suite spanning reasoning, agentic, general, Korean, and multilingual abilities. Across these evaluations, K-EXAONE demonstrates performance comparable to open-weight models of similar size. K-EXAONE, designed to advance AI for a better life, is positioned as a powerful proprietary AI foundation model for a wide range of industrial and research applications.

2601.01739

Jan 2026Computation and Language

Extracting books from production language models

Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model's weights during training, and whether those memorized data can be extracted in the model's outputs. While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models. However, it remains an open question if similar extraction is feasible for production LLMs, given the safety measures these systems implement. We investigate this question using a two-phase procedure: (1) an initial probe to test for extraction feasibility, which sometimes uses a Best-of-N (BoN) jailbreak, followed by (2) iterative continuation prompts to attempt to extract the book. We evaluate our procedure on four production LLMs -- Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3 -- and we measure extraction success with a score computed from a block-based approximation of longest common substring (nv-recall). With different per-LLM experimental configurations, we were able to extract varying amounts of text. For the Phase 1 probe, it was unnecessary to jailbreak Gemini 2.5 Pro and Grok 3 to extract text (e.g, nv-recall of 76.8% and 70.3%, respectively, for Harry Potter and the Sorcerer's Stone), while it was necessary for Claude 3.7 Sonnet and GPT-4.1. In some cases, jailbroken Claude 3.7 Sonnet outputs entire books near-verbatim (e.g., nv-recall=95.8%). GPT-4.1 requires significantly more BoN attempts (e.g., 20X), and eventually refuses to continue (e.g., nv-recall=4.0%). Taken together, our work highlights that, even with model- and system-level safeguards, extraction of (in-copyright) training data remains a risk for production LLMs.

2601.02671

Jan 2026Computation and Language

From Failure to Mastery: Generating Hard Samples for Tool-use Agents

The advancement of LLM agents with tool-use capabilities requires diverse and complex training corpora. Existing data generation methods, which predominantly follow a paradigm of random sampling and shallow generation, often yield simple and homogeneous trajectories that fail to capture complex, implicit logical dependencies. To bridge this gap, we introduce HardGen, an automatic agentic pipeline designed to generate hard tool-use training samples with verifiable reasoning. Firstly, HardGen establishes a dynamic API Graph built upon agent failure cases, from which it samples to synthesize hard traces. Secondly, these traces serve as conditional priors to guide the instantiation of modular, abstract advanced tools, which are subsequently leveraged to formulate hard queries. Finally, the advanced tools and hard queries enable the generation of verifiable complex Chain-of-Thought (CoT), with a closed-loop evaluation feedback steering the continuous refinement of the process. Extensive evaluations demonstrate that a 4B parameter model trained with our curated dataset achieves superior performance compared to several leading open-source and closed-source competitors (e.g., GPT-5.2, Gemini-3-Pro and Claude-Opus-4.5). Our code, models, and dataset will be open-sourced to facilitate future research.

2601.01498

Jan 2026Computation and Language

WebAnchor: Anchoring Agent Planning to Stabilize Long-Horizon Web Reasoning

Large Language Model(LLM)-based agents have shown strong capabilities in web information seeking, with reinforcement learning (RL) becoming a key optimization paradigm. However, planning remains a bottleneck, as existing methods struggle with long-horizon strategies. Our analysis reveals a critical phenomenon, plan anchor, where the first reasoning step disproportionately impacts downstream behavior in long-horizon web reasoning tasks. Current RL algorithms, fail to account for this by uniformly distributing rewards across the trajectory. To address this, we propose Anchor-GRPO, a two-stage RL framework that decouples planning and execution. In Stage 1, the agent optimizes its first-step planning using fine-grained rubrics derived from self-play experiences and human calibration. In Stage 2, execution is aligned with the initial plan through sparse rewards, ensuring stable and efficient tool usage. We evaluate Anchor-GRPO on four benchmarks: BrowseComp, BrowseComp-Zh, GAIA, and XBench-DeepSearch. Across models from 3B to 30B, Anchor-GRPO outperforms baseline GRPO and First-step GRPO, improving task success and tool efficiency. Notably, WebAnchor-30B achieves 46.0% pass@1 on BrowseComp and 76.4% on GAIA. Anchor-GRPO also demonstrates strong scalability, getting higher accuracy as model size and context length increase.

2601.03164

Jan 2026Computation and Language

MiMo-V2-Flash Technical Report

We present MiMo-V2-Flash, a Mixture-of-Experts (MoE) model with 309B total parameters and 15B active parameters, designed for fast, strong reasoning and agentic capabilities. MiMo-V2-Flash adopts a hybrid attention architecture that interleaves Sliding Window Attention (SWA) with global attention, with a 128-token sliding window under a 5:1 hybrid ratio. The model is pre-trained on 27 trillion tokens with Multi-Token Prediction (MTP), employing a native 32k context length and subsequently extended to 256k. To efficiently scale post-training compute, MiMo-V2-Flash introduces a novel Multi-Teacher On-Policy Distillation (MOPD) paradigm. In this framework, domain-specialized teachers (e.g., trained via large-scale reinforcement learning) provide dense and token-level reward, enabling the student model to perfectly master teacher expertise. MiMo-V2-Flash rivals top-tier open-weight models such as DeepSeek-V3.2 and Kimi-K2, despite using only 1/2 and 1/3 of their total parameters, respectively. During inference, by repurposing MTP as a draft model for speculative decoding, MiMo-V2-Flash achieves up to 3.6 acceptance length and 2.6x decoding speedup with three MTP layers. We open-source both the model weights and the three-layer MTP weights to foster open research and community collaboration.

2601.02780

Jan 2026Computation and Language

Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning

The integration of large language models (LLMs) with external tools has significantly expanded the capabilities of AI agents. However, as the diversity of both LLMs and tools increases, selecting the optimal model-tool combination becomes a high-dimensional optimization challenge. Existing approaches often rely on a single model or fixed tool-calling logic, failing to exploit the performance variations across heterogeneous model-tool pairs. In this paper, we present ATLAS (Adaptive Tool-LLM Alignment and Synergistic Invocation), a dual-path framework for dynamic tool usage in cross-domain complex reasoning. ATLAS operates via a dual-path approach: (1) \textbf{training-free cluster-based routing} that exploits empirical priors for domain-specific alignment, and (2) \textbf{RL-based multi-step routing} that explores autonomous trajectories for out-of-distribution generalization. Extensive experiments across 15 benchmarks demonstrate that our method outperforms closed-source models like GPT-4o, surpassing existing routing methods on both in-distribution (+10.1%) and out-of-distribution (+13.1%) tasks. Furthermore, our framework shows significant gains in visual reasoning by orchestrating specialized multi-modal tools.

2601.03872

Jan 2026Computation and Language

MedDialogRubrics: A Comprehensive Benchmark and Evaluation Framework for Multi-turn Medical Consultations in Large Language Models

Medical conversational AI (AI) plays a pivotal role in the development of safer and more effective medical dialogue systems. However, existing benchmarks and evaluation frameworks for assessing the information-gathering and diagnostic reasoning abilities of medical large language models (LLMs) have not been rigorously evaluated. To address these gaps, we present MedDialogRubrics, a novel benchmark comprising 5,200 synthetically constructed patient cases and over 60,000 fine-grained evaluation rubrics generated by LLMs and subsequently refined by clinical experts, specifically designed to assess the multi-turn diagnostic capabilities of LLM. Our framework employs a multi-agent system to synthesize realistic patient records and chief complaints from underlying disease knowledge without accessing real-world electronic health records, thereby mitigating privacy and data-governance concerns. We design a robust Patient Agent that is limited to a set of atomic medical facts and augmented with a dynamic guidance mechanism that continuously detects and corrects hallucinations throughout the dialogue, ensuring internal coherence and clinical plausibility of the simulated cases. Furthermore, we propose a structured LLM-based and expert-annotated rubric-generation pipeline that retrieves Evidence-Based Medicine (EBM) guidelines and utilizes the reject sampling to derive a prioritized set of rubric items ("must-ask" items) for each case. We perform a comprehensive evaluation of state-of-the-art models and demonstrate that, across multiple assessment dimensions, current models face substantial challenges. Our results indicate that improving medical dialogue will require advances in dialogue management architectures, not just incremental tuning of the base-model.

2601.03023

Jan 2026Computation and Language

Visual Merit or Linguistic Crutch? A Close Look at DeepSeek-OCR

DeepSeek-OCR utilizes an optical 2D mapping approach to achieve high-ratio vision-text compression, claiming to decode text tokens exceeding ten times the input visual tokens. While this suggests a promising solution for the LLM long-context bottleneck, we investigate a critical question: "Visual merit or linguistic crutch - which drives DeepSeek-OCR's performance?" By employing sentence-level and word-level semantic corruption, we isolate the model's intrinsic OCR capabilities from its language priors. Results demonstrate that without linguistic support, DeepSeek-OCR's performance plummets from approximately 90% to 20%. Comparative benchmarking against 13 baseline models reveals that traditional pipeline OCR methods exhibit significantly higher robustness to such semantic perturbations than end-to-end methods. Furthermore, we find that lower visual token counts correlate with increased reliance on priors, exacerbating hallucination risks. Context stress testing also reveals a total model collapse around 10,000 text tokens, suggesting that current optical compression techniques may paradoxically aggravate the long-context bottleneck. This study empirically defines DeepSeek-OCR's capability boundaries and offers essential insights for future optimizations of the vision-text compression paradigm. We release all data, results and scripts used in this study at https://github.com/dududuck00/DeepSeekOCR.

2601.03714

Jan 2026Computation and Language

InfiniteWeb: Scalable Web Environment Synthesis for GUI Agent Training

GUI agents that interact with graphical interfaces on behalf of users represent a promising direction for practical AI assistants. However, training such agents is hindered by the scarcity of suitable environments. We present InfiniteWeb, a system that automatically generates functional web environments at scale for GUI agent training. While LLMs perform well on generating a single webpage, building a realistic and functional website with many interconnected pages faces challenges. We address these challenges through unified specification, task-centric test-driven development, and a combination of website seed with reference design image to ensure diversity. Our system also generates verifiable task evaluators enabling dense reward signals for reinforcement learning. Experiments show that InfiniteWeb surpasses commercial coding agents at realistic website construction, and GUI agents trained on our generated environments achieve significant performance improvements on OSWorld and Online-Mind2Web, demonstrating the effectiveness of proposed system.

2601.04126

Jan 2026Computation and Language

Mechanistic Interpretability of Large-Scale Counting in LLMs through a System-2 Strategy

Large language models (LLMs), despite strong performance on complex mathematical problems, exhibit systematic limitations in counting tasks. This issue arises from architectural limits of transformers, where counting is performed across layers, leading to degraded precision for larger counting problems due to depth constraints. To address this limitation, we propose a simple test-time strategy inspired by System-2 cognitive processes that decomposes large counting tasks into smaller, independent sub-problems that the model can reliably solve. We evaluate this approach using observational and causal mediation analyses to understand the underlying mechanism of this System-2-like strategy. Our mechanistic analysis identifies key components: latent counts are computed and stored in the final item representations of each part, transferred to intermediate steps via dedicated attention heads, and aggregated in the final stage to produce the total count. Experimental results demonstrate that this strategy enables LLMs to surpass architectural limitations and achieve high accuracy on large-scale counting tasks. This work provides mechanistic insight into System-2 counting in LLMs and presents a generalizable approach for improving and understanding their reasoning behavior.

2601.02989

Jan 2026Computation and Language

Benchmark^2: Systematic Evaluation of LLM Benchmarks

The rapid proliferation of benchmarks for evaluating large language models (LLMs) has created an urgent need for systematic methods to assess benchmark quality itself. We propose Benchmark^2, a comprehensive framework comprising three complementary metrics: (1) Cross-Benchmark Ranking Consistency, measuring whether a benchmark produces model rankings aligned with peer benchmarks; (2) Discriminability Score, quantifying a benchmark's ability to differentiate between models; and (3) Capability Alignment Deviation, identifying problematic instances where stronger models fail but weaker models succeed within the same model family. We conduct extensive experiments across 15 benchmarks spanning mathematics, reasoning, and knowledge domains, evaluating 11 LLMs across four model families. Our analysis reveals significant quality variations among existing benchmarks and demonstrates that selective benchmark construction based on our metrics can achieve comparable evaluation performance with substantially reduced test sets.

2601.03986

Jan 2026Computation and Language

TiMem: Temporal-Hierarchical Memory Consolidation for Long-Horizon Conversational Agents

Long-horizon conversational agents have to manage ever-growing interaction histories that quickly exceed the finite context windows of large language models (LLMs). Existing memory frameworks provide limited support for temporally structured information across hierarchical levels, often leading to fragmented memories and unstable long-horizon personalization. We present TiMem, a temporal--hierarchical memory framework that organizes conversations through a Temporal Memory Tree (TMT), enabling systematic memory consolidation from raw conversational observations to progressively abstracted persona representations. TiMem is characterized by three core properties: (1) temporal--hierarchical organization through TMT; (2) semantic-guided consolidation that enables memory integration across hierarchical levels without fine-tuning; and (3) complexity-aware memory recall that balances precision and efficiency across queries of varying complexity. Under a consistent evaluation setup, TiMem achieves state-of-the-art accuracy on both benchmarks, reaching 75.30% on LoCoMo and 76.88% on LongMemEval-S. It outperforms all evaluated baselines while reducing the recalled memory length by 52.20% on LoCoMo. Manifold analysis indicates clear persona separation on LoCoMo and reduced dispersion on LongMemEval-S. Overall, TiMem treats temporal continuity as a first-class organizing principle for long-horizon memory in conversational agents.

2601.02845

Jan 2026Computation and Language

pdfQA: Diverse, Challenging, and Realistic Question Answering over PDFs

PDFs are the second-most used document type on the internet (after HTML). Yet, existing QA datasets commonly start from text sources or only address specific domains. In this paper, we present pdfQA, a multi-domain 2K human-annotated (real-pdfQA) and 2K synthetic dataset (syn-pdfQA) differentiating QA pairs in ten complexity dimensions (e.g., file type, source modality, source position, answer type). We apply and evaluate quality and difficulty filters on both datasets, obtaining valid and challenging QA pairs. We answer the questions with open-source LLMs, revealing existing challenges that correlate with our complexity dimensions. pdfQA presents a basis for end-to-end QA pipeline evaluation, testing diverse skill sets and local optimizations (e.g., in information retrieval or parsing).

2601.02285

Jan 2026Computation and Language

O-Researcher: An Open Ended Deep Research Model via Multi-Agent Distillation and Agentic RL

The performance gap between closed-source and open-source large language models (LLMs) is largely attributed to disparities in access to high-quality training data. To bridge this gap, we introduce a novel framework for the automated synthesis of sophisticated, research-grade instructional data. Our approach centers on a multi-agent workflow where collaborative AI agents simulate complex tool-integrated reasoning to generate diverse and high-fidelity data end-to-end. Leveraging this synthesized data, we develop a two-stage training strategy that integrates supervised fine-tuning with a novel reinforcement learning method, designed to maximize model alignment and capability. Extensive experiments demonstrate that our framework empowers open-source models across multiple scales, enabling them to achieve new state-of-the-art performance on the major deep research benchmark. This work provides a scalable and effective pathway for advancing open-source LLMs without relying on proprietary data or models.

2601.03743

Jan 2026Computation and Language

DIP: Dynamic In-Context Planner For Diffusion Language Models

Diffusion language models (DLMs) have shown strong potential for general natural language tasks with in-context examples. However, due to the bidirectional attention mechanism, DLMs incur substantial computational cost as context length increases. This work addresses this issue with a key discovery: unlike the sequential generation in autoregressive language models (ARLMs), the diffusion generation paradigm in DLMs allows \textit{efficient dynamic adjustment of the context} during generation. Building on this insight, we propose \textbf{D}ynamic \textbf{I}n-Context \textbf{P}lanner (DIP), a context-optimization method that dynamically selects and inserts in-context examples during generation, rather than providing all examples in the prompt upfront. Results show DIP maintains generation quality while achieving up to 12.9$\times$ inference speedup over standard inference and 1.17$\times$ over KV cache-enhanced inference.

2601.03199

Jan 2026Computation and Language

Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models

We introduce Youtu-LLM, a lightweight yet powerful language model that harmonizes high computational efficiency with native agentic intelligence. Unlike typical small models that rely on distillation, Youtu-LLM (1.96B) is pre-trained from scratch to systematically cultivate reasoning and planning capabilities. The key technical advancements are as follows: (1) Compact Architecture with Long-Context Support: Built on a dense Multi-Latent Attention (MLA) architecture with a novel STEM-oriented vocabulary, Youtu-LLM supports a 128k context window. This design enables robust long-context reasoning and state tracking within a minimal memory footprint, making it ideal for long-horizon agent and reasoning tasks. (2) Principled "Commonsense-STEM-Agent" Curriculum: We curated a massive corpus of approximately 11T tokens and implemented a multi-stage training strategy. By progressively shifting the pre-training data distribution from general commonsense to complex STEM and agentic tasks, we ensure the model acquires deep cognitive abilities rather than superficial alignment. (3) Scalable Agentic Mid-training: Specifically for the agentic mid-training, we employ diverse data construction schemes to synthesize rich and varied trajectories across math, coding, and tool-use domains. This high-quality data enables the model to internalize planning and reflection behaviors effectively. Extensive evaluations show that Youtu-LLM sets a new state-of-the-art for sub-2B LLMs. On general benchmarks, it achieves competitive performance against larger models, while on agent-specific tasks, it significantly surpasses existing SOTA baselines, demonstrating that lightweight models can possess strong intrinsic agentic capabilities.

2512.24618

Dec 2025Computation and Language

12,965 papers

All That Glisters Is Not Gold: A Benchmark for Reference-Free Counterfactual Financial Misinformation Detection

We introduce RFC Bench, a benchmark for evaluating large language models on financial misinformation under realistic news. RFC Bench operates at the paragraph level and captures the contextual complexity of financial news where meaning emerges from dispersed cues. The benchmark defines two complementary tasks: reference free misinformation detection and comparison based diagnosis using paired original perturbed inputs. Experiments reveal a consistent pattern: performance is substantially stronger when comparative context is available, while reference free settings expose significant weaknesses, including unstable predictions and elevated invalid outputs. These results indicate that current models struggle to maintain coherent belief states without external grounding. By highlighting this gap, RFC Bench provides a structured testbed for studying reference free reasoning and advancing more reliable financial misinformation detection in real world settings.

2601.04160Jan 2026

View

FLEx: Language Modeling with Few-shot Language Explanations

Language models have become effective at a wide range of tasks, from math problem solving to open-domain question answering. However, they still make mistakes, and these mistakes are often repeated across related queries. Natural language explanations can help correct these errors, but collecting them at scale may be infeasible, particularly in domains where expert annotators are required. To address this issue, we introduce FLEx ($\textbf{F}$ew-shot $\textbf{L}$anguage $\textbf{Ex}$planations), a method for improving model behavior using a small number of explanatory examples. FLEx selects representative model errors using embedding-based clustering, verifies that the associated explanations correct those errors, and summarizes them into a prompt prefix that is prepended at inference-time. This summary guides the model to avoid similar errors on new inputs, without modifying model weights. We evaluate FLEx on CounterBench, GSM8K, and ReasonIF. We find that FLEx consistently outperforms chain-of-thought (CoT) prompting across all three datasets and reduces up to 83\% of CoT's remaining errors.

2601.04157Jan 2026

View

LLMberjack: Guided Trimming of Debate Trees for Multi-Party Conversation Creation

We present LLMberjack, a platform for creating multi-party conversations starting from existing debates, originally structured as reply trees. The system offers an interactive interface that visualizes discussion trees and enables users to construct coherent linearized dialogue sequences while preserving participant identity and discourse relations. It integrates optional large language model (LLM) assistance to support automatic editing of the messages and speakers' descriptions. We demonstrate the platform's utility by showing how tree visualization facilitates the creation of coherent, meaningful conversation threads and how LLM support enhances output quality while reducing human effort. The tool is open-source and designed to promote transparent and reproducible workflows to create multi-party conversations, addressing a lack of resources of this type.

2601.04135Jan 2026

View

ContextFocus: Activation Steering for Contextual Faithfulness in Large Language Models

Large Language Models (LLMs) encode vast amounts of parametric knowledge during pre-training. As world knowledge evolves, effective deployment increasingly depends on their ability to faithfully follow externally retrieved context. When such evidence conflicts with the model's internal knowledge, LLMs often default to memorized facts, producing unfaithful outputs. In this work, we introduce ContextFocus, a lightweight activation steering approach that improves context faithfulness in such knowledge-conflict settings while preserving fluency and efficiency. Unlike prior approaches, our solution requires no model finetuning and incurs minimal inference-time overhead, making it highly efficient. We evaluate ContextFocus on the ConFiQA benchmark, comparing it against strong baselines including ContextDPO, COIECD, and prompting-based methods. Furthermore, we show that our method is complementary to prompting strategies and remains effective on larger models. Extensive experiments show that ContextFocus significantly improves contextual-faithfulness. Our results highlight the effectiveness, robustness, and efficiency of ContextFocus in improving contextual-faithfulness of LLM outputs.

2601.04131Jan 2026

View

InfiniteWeb: Scalable Web Environment Synthesis for GUI Agent Training

2601.04126Jan 2026

View

Layer-wise Positional Bias in Short-Context Language Modeling

Language models often show a preference for using information from specific positions in the input regardless of semantic relevance. While positional bias has been studied in various contexts, from attention sinks to task performance degradation in long-context settings, prior work has not established how these biases evolve across individual layers and input positions, or how they vary independent of task complexity. We introduce an attribution-based framework to analyze positional effects in short-context language modeling. Using layer conductance with a sliding-window approach, we quantify how each layer distributes importance across input positions, yielding layer-wise positional importance profiles. We find that these profiles are architecture-specific, stable across inputs, and invariant to lexical scrambling. Characterizing these profiles, we find prominent recency bias that increases with depth and subtle primacy bias that diminishes through model depth. Beyond positional structure, we also show that early layers preferentially weight content words over function words across all positions, while later layers lose this word-type differentiation.

2601.04098Jan 2026

View

SearchAttack: Red-Teaming LLMs against Real-World Threats via Framing Unsafe Web Information-Seeking Tasks

Recently, people have suffered and become increasingly aware of the unreliability gap in LLMs for open and knowledge-intensive tasks, and thus turn to search-augmented LLMs to mitigate this issue. However, when the search engine is triggered for harmful tasks, the outcome is no longer under the LLM's control. Once the returned content directly contains targeted, ready-to-use harmful takeaways, the LLM's safeguards cannot withdraw that exposure. Motivated by this dilemma, we identify web search as a critical attack surface and propose \textbf{\textit{SearchAttack}} for red-teaming. SearchAttack outsources the harmful semantics to web search, retaining only the query's skeleton and fragmented clues, and further steers LLMs to reconstruct the retrieved content via structural rubrics to achieve malicious goals. Extensive experiments are conducted to red-team the search-augmented LLMs for responsible vulnerability assessment. Empirically, SearchAttack demonstrates strong effectiveness in attacking these systems.

2601.04093Jan 2026

View

KDCM: Reducing Hallucination in LLMs through Explicit Reasoning Structures

To mitigate hallucinations in large language models (LLMs), we propose a framework that focuses on errors induced by prompts. Our method extends a chain-style knowledge distillation approach by incorporating a programmable module that guides knowledge graph exploration. This module is embedded as executable code within the reasoning prompt, allowing the model to leverage external structured knowledge during inference. Based on this design, we develop an enhanced distillation-based reasoning framework that explicitly regulates intermediate reasoning steps, resulting in more reliable predictions. We evaluate the proposed approach on multiple public benchmarks using GPT-4 and LLaMA-3.3. Experimental results show that code-guided reasoning significantly improves contextual modeling and reduces prompt-induced hallucinations. Specifically, HIT@1, HIT@3, and HIT@5 increase by 15.64%, 13.38%, and 13.28%, respectively, with scores exceeding 95% across several evaluation settings. These findings indicate that the proposed method effectively constrains erroneous reasoning while improving both accuracy and interpretability.

2601.04086Jan 2026

View

Bridging the Discrete-Continuous Gap: Unified Multimodal Generation via Coupled Manifold Discrete Absorbing Diffusion

The bifurcation of generative modeling into autoregressive approaches for discrete data (text) and diffusion approaches for continuous data (images) hinders the development of truly unified multimodal systems. While Masked Language Models (MLMs) offer efficient bidirectional context, they traditionally lack the generative fidelity of autoregressive models and the semantic continuity of diffusion models. Furthermore, extending masked generation to multimodal settings introduces severe alignment challenges and training instability. In this work, we propose \textbf{CoM-DAD} (\textbf{Co}upled \textbf{M}anifold \textbf{D}iscrete \textbf{A}bsorbing \textbf{D}iffusion), a novel probabilistic framework that reformulates multimodal generation as a hierarchical dual-process. CoM-DAD decouples high-level semantic planning from low-level token synthesis. First, we model the semantic manifold via a continuous latent diffusion process; second, we treat token generation as a discrete absorbing diffusion process, regulated by a \textbf{Variable-Rate Noise Schedule}, conditioned on these evolving semantic priors. Crucially, we introduce a \textbf{Stochastic Mixed-Modal Transport} strategy that aligns disparate modalities without requiring heavy contrastive dual-encoders. Our method demonstrates superior stability over standard masked modeling, establishing a new paradigm for scalable, unified text-image generation.

2601.04056Jan 2026

View

Modular Prompt Optimization: Optimizing Structured Prompts with Section-Local Textual Gradients

Prompt quality plays a central role in controlling the behavior, reliability, and reasoning performance of large language models (LLMs), particularly for smaller open-source instruction-tuned models that depend heavily on explicit structure. While recent work has explored automatic prompt optimization using textual gradients and self-refinement, most existing methods treat prompts as monolithic blocks of text, making it difficult to localize errors, preserve critical instructions, or prevent uncontrolled prompt growth. We introduce Modular Prompt Optimization (MPO), a schema-based prompt optimization framework that treats prompts as structured objects composed of fixed semantic sections, including system role, context, task description, constraints, and output format. MPO applies section-local textual gradients, generated by a critic language model, to refine each section independently while keeping the overall prompt schema fixed. Section updates are consolidated through de-duplication to reduce redundancy and interference between components, yielding an interpretable and robust optimization process. We evaluate MPO on two reasoning benchmarks, ARC-Challenge and MMLU, using LLaMA-3 8B-Instruct and Mistral-7B-Instruct as solver models. Across both benchmarks and models, MPO consistently outperforms an untuned structured prompt and the TextGrad baseline, achieving substantial accuracy gains without modifying model parameters or altering prompt structure. These results demonstrate that maintaining a fixed prompt schema while applying localized, section-wise optimization is an effective and practical approach for improving reasoning performance in small open-source LMs.

2601.04055Jan 2026

View

When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Life

As Multimodal Large Language Models (MLLMs) become an indispensable assistant in human life, the unsafe content generated by MLLMs poses a danger to human behavior, perpetually overhanging human society like a sword of Damocles. To investigate and evaluate the safety impact of MLLMs responses on human behavior in daily life, we introduce SaLAD, a multimodal safety benchmark which contains 2,013 real-world image-text samples across 10 common categories, with a balanced design covering both unsafe scenarios and cases of oversensitivity. It emphasizes realistic risk exposure, authentic visual inputs, and fine-grained cross-modal reasoning, ensuring that safety risks cannot be inferred from text alone. We further propose a safety-warning-based evaluation framework that encourages models to provide clear and informative safety warnings, rather than generic refusals. Results on 18 MLLMs demonstrate that the top-performing models achieve a safe response rate of only 57.2% on unsafe queries. Moreover, even popular safety alignment methods limit effectiveness of the models in our scenario, revealing the vulnerabilities of current MLLMs in identifying dangerous behaviors in daily life. Our dataset is available at https://github.com/xinyuelou/SaLAD.

2601.04043Jan 2026

View

Analyzing and Improving Cross-lingual Knowledge Transfer for Machine Translation

Multilingual machine translation systems aim to make knowledge accessible across languages, yet learning effective cross-lingual representations remains challenging. These challenges are especially pronounced for low-resource languages, where limited parallel data constrains generalization and transfer. Understanding how multilingual models share knowledge across languages requires examining the interaction between representations, data availability, and training strategies. In this thesis, we study cross-lingual knowledge transfer in neural models and develop methods to improve robustness and generalization in multilingual settings, using machine translation as a central testbed. We analyze how similarity between languages influences transfer, how retrieval and auxiliary supervision can strengthen low-resource translation, and how fine-tuning on parallel data can introduce unintended trade-offs in large language models. We further examine the role of language diversity during training and show that increasing translation coverage improves generalization and reduces off-target behavior. Together, this work highlights how modeling choices and data composition shape multilingual learning and offers insights toward more inclusive and resilient multilingual NLP systems.

2601.04036Jan 2026

View

SpeakerSleuth: Evaluating Large Audio-Language Models as Judges for Multi-turn Speaker Consistency

Large Audio-Language Models (LALMs) as judges have emerged as a prominent approach for evaluating speech generation quality, yet their ability to assess speaker consistency across multi-turn conversations remains unexplored. We present SpeakerSleuth, a benchmark evaluating whether LALMs can reliably judge speaker consistency in multi-turn dialogues through three tasks reflecting real-world requirements. We construct 1,818 human-verified evaluation instances across four diverse datasets spanning synthetic and real speech, with controlled acoustic difficulty. Evaluating nine widely-used LALMs, we find that models struggle to reliably detect acoustic inconsistencies. For instance, given audio samples of the same speaker's turns, some models overpredict inconsistency, whereas others are overly lenient. Models further struggle to identify the exact turns that are problematic. When other interlocutors' turns are provided together, performance degrades dramatically as models prioritize textual coherence over acoustic cues, failing to detect even obvious gender switches for a speaker. On the other hand, models perform substantially better in choosing the audio that best matches the speaker among several acoustic variants, demonstrating inherent acoustic discrimination capabilities. These findings expose a significant bias in LALMs: they tend to prioritize text over acoustics, revealing fundamental modality imbalances that need to be addressed to build reliable audio-language judges.

2601.04029Jan 2026

View

Simulated Students in Tutoring Dialogues: Substance or Illusion?

Advances in large language models (LLMs) enable many new innovations in education. However, evaluating the effectiveness of new technology requires real students, which is time-consuming and hard to scale up. Therefore, many recent works on LLM-powered tutoring solutions have used simulated students for both training and evaluation, often via simple prompting. Surprisingly, little work has been done to ensure or even measure the quality of simulated students. In this work, we formally define the student simulation task, propose a set of evaluation metrics that span linguistic, behavioral, and cognitive aspects, and benchmark a wide range of student simulation methods on these metrics. We experiment on a real-world math tutoring dialogue dataset, where both automated and human evaluation results show that prompting strategies for student simulation perform poorly; supervised fine-tuning and preference optimization yield much better but still limited performance, motivating future work on this challenging task.

2601.04025Jan 2026

View

2601.03997

VotIE: Information Extraction from Meeting Minutes

Municipal meeting minutes record key decisions in local democratic processes. Unlike parliamentary proceedings, which typically adhere to standardized formats, they encode voting outcomes in highly heterogeneous, free-form narrative text that varies widely across municipalities, posing significant challenges for automated extraction. In this paper, we introduce VotIE (Voting Information Extraction), a new information extraction task aimed at identifying structured voting events in narrative deliberative records, and establish the first benchmark for this task using Portuguese municipal minutes, building on the recently introduced CitiLink corpus. Our experiments yield two key findings. First, under standard in-domain evaluation, fine-tuned encoders, specifically XLM-R-CRF, achieve the strongest performance, reaching 93.2\% macro F1, outperforming generative approaches. Second, in a cross-municipality setting that evaluates transfer to unseen administrative contexts, these models suffer substantial performance degradation, whereas few-shot LLMs demonstrate greater robustness, with significantly smaller declines in performance. Despite this generalization advantage, the high computational cost of generative models currently constrains their practicality. As a result, lightweight fine-tuned encoders remain a more practical option for large-scale, real-world deployment. To support reproducible research in administrative NLP, we publicly release our benchmark, trained models, and evaluation framework.

2601.03997Jan 2026

View

Benchmark^2: Systematic Evaluation of LLM Benchmarks

2601.03986Jan 2026

View

RADAR: Retrieval-Augmented Detector with Adversarial Refinement for Robust Fake News Detection

To efficiently combat the spread of LLM-generated misinformation, we present RADAR, a retrieval-augmented detector with adversarial refinement for robust fake news detection. Our approach employs a generator that rewrites real articles with factual perturbations, paired with a lightweight detector that verifies claims using dense passage retrieval. To enable effective co-evolution, we introduce verbal adversarial feedback (VAF). Rather than relying on scalar rewards, VAF issues structured natural-language critiques; these guide the generator toward more sophisticated evasion attempts, compelling the detector to adapt and improve. On a fake news detection benchmark, RADAR achieves 86.98% ROC-AUC, significantly outperforming general-purpose LLMs with retrieval. Ablation studies confirm that detector-side retrieval yields the largest gains, while VAF and few-shot demonstrations provide critical signals for robust training.

2601.03981Jan 2026

View

Large-Scale Aspect-Based Sentiment Analysis with Reasoning-Infused LLMs

We introduce Arctic-ABSA, a collection of powerful models for real-life aspect-based sentiment analysis (ABSA). Our models are tailored to commercial needs, trained on a large corpus of public data alongside carefully generated synthetic data, resulting in a dataset 20 times larger than SemEval14. We extend typical ABSA models by expanding the number of sentiment classes from the standard three (positive, negative, neutral) to five, adding mixed and unknown classes, while also jointly predicting overall text sentiment and supporting multiple languages. We experiment with reasoning injection by fine-tuning on Chain-of-Thought (CoT) examples and introduce a novel reasoning pretraining technique for encoder-only models that significantly improves downstream fine-tuning and generalization. Our 395M-parameter encoder and 8B-parameter decoder achieve up to 10 percentage points higher accuracy than GPT-4o and Claude 3.5 Sonnet, while setting new state-of-the-art results on the SemEval14 benchmark. A single multilingual model maintains 87-91% accuracy across six languages without degrading English performance. We release ABSA-mix, a large-scale benchmark aggregating 17 public ABSA datasets across 92 domains.

2601.03940Jan 2026

View

Doc-PP: Document Policy Preservation Benchmark for Large Vision-Language Models

The deployment of Large Vision-Language Models (LVLMs) for real-world document question answering is often constrained by dynamic, user-defined policies that dictate information disclosure based on context. While ensuring adherence to these explicit constraints is critical, existing safety research primarily focuses on implicit social norms or text-only settings, overlooking the complexities of multimodal documents. In this paper, we introduce Doc-PP (Document Policy Preservation Benchmark), a novel benchmark constructed from real-world reports requiring reasoning across heterogeneous visual and textual elements under strict non-disclosure policies. Our evaluation highlights a systemic Reasoning-Induced Safety Gap: models frequently leak sensitive information when answers must be inferred through complex synthesis or aggregated across modalities, effectively circumventing existing safety constraints. Furthermore, we identify that providing extracted text improves perception but inadvertently facilitates leakage. To address these vulnerabilities, we propose DVA (Decompose-Verify-Aggregation), a structural inference framework that decouples reasoning from policy verification. Experimental results demonstrate that DVA significantly outperforms standard prompting defenses, offering a robust baseline for policy-compliant document understanding

2601.03926Jan 2026

View

When Models Decide and When They Bind: A Two-Stage Computation for Multiple-Choice Question-Answering

Multiple-choice question answering (MCQA) is easy to evaluate but adds a meta-task: models must both solve the problem and output the symbol that *represents* the answer, conflating reasoning errors with symbol-binding failures. We study how language models implement MCQA internally using representational analyses (PCA, linear probes) as well as causal interventions. We find that option-boundary (newline) residual states often contain strong linearly decodable signals related to per-option correctness. Winner-identity probing reveals a two-stage progression: the winning *content position* becomes decodable immediately after the final option is processed, while the *output symbol* is represented closer to the answer emission position. Tests under symbol and content permutations support a two-stage mechanism in which models first select a winner in content space and then bind or route that winner to the appropriate symbol to emit.

2601.03914Jan 2026

View

Page 1 of 649