Peer Revenue Impact factor: volatile

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Monday 20 October

Computation and Language

Na Prática, qual IA Entende o Direito? Um Estudo Experimental com IAs Generalistas e uma IA Jurídica

Marina Soares Marinho, Daniela Vianna, Livy Real et al. (+2 more)

This study presents the Jusbrasil Study on the Use of General-Purpose AIs in Law, proposing an experimental evaluation protocol combining legal theory, such as material correctness, systematic coherence, and argumentative integrity, with empirical assessment by 48 legal professionals. Four systems (JusIA, ChatGPT Free, ChatGPT Pro, and Gemini) were tested in tasks simulating lawyers' daily work. JusIA, a domain-specialized model, consistently outperformed the general-purpose systems, showing that both domain specialization and a theoretically grounded evaluation are essential for reliable legal AI outputs.
Read the full abstract

All authors Marina Soares Marinho, Daniela Vianna, Livy Real, Altigran da Silva, Gabriela Migliorini

Computation and Language · Cross-listed from cs.AI

SMaRT: Select, Mix, and ReinvenT – A Strategy Fusion Framework for LLM-Driven Reasoning and Planning

Nikhil Verma, Manasa Bharadwaj, Wonjun Jang et al. (+4 more)

Large Language Models (LLMs) have redefined complex task automation with exceptional generalization capabilities. Despite these advancements, state-of-the-art methods rely on single-strategy prompting, missing the synergy of diverse reasoning approaches. No single strategy excels universally, highlighting the need for frameworks that fuse strategies to maximize performance and ensure robustness. We introduce the Select, Mix, and ReinvenT (SMaRT) framework, an innovative strategy fusion approach designed to overcome this constraint by creating balanced and efficient solutions through the seamless integration of diverse reasoning strategies. Unlike existing methods, which employ LLMs merely as evaluators, SMaRT uses them as intelligent integrators, unlocking the "best of all worlds" across tasks. Extensive empirical evaluations across benchmarks in reasoning, planning, and sequential decision-making highlight the robustness and adaptability of SMaRT. The framework consistently outperforms state-of-the-art baselines in solution quality, constraint adherence, and performance metrics. This work redefines LLM-driven decision-making by pioneering a new paradigm in cross-strategy calibration, unlocking superior outcomes for reasoning systems and advancing the boundaries of self-refining methodologies.
Read the full abstract

All authors Nikhil Verma, Manasa Bharadwaj, Wonjun Jang, Harmanpreet Singh, Yixiao Wang, Homa Fashandi, Chul Lee

Computation and Language

Chain-of-Thought Reasoning Improves Context-Aware Translation with Large Language Models

Shabnam Ataee, Hugo Huart and Andrei Popescu-Belis

This paper assesses the ability of large language models (LLMs) to translate texts that include inter-sentential dependencies. We use the English-French DiscEvalMT benchmark (Bawden et al., 2018) with pairs of sentences containing translation challenges for pronominal anaphora and lexical cohesion. We evaluate 12 LLMs from the DeepSeek-R1, GPT, Llama, Mistral and Phi families on two tasks: (1) distinguish a correct translation from a wrong but plausible one; and (2) generate a correct translation. We compare prompts that encourage chain-of-thought reasoning with those that do not. The best models take advantage of reasoning and reach about 90% accuracy on the first task and COMET scores of about 92% on the second task, with GPT-4, GPT-4o and Phi standing out. Moreover, we observe a "wise get wiser" effect: the improvements through reasoning are larger for models that already perform well without reasoning.
Read the full abstract

Computation and Language · Cross-listed from cs.CV

HouseTour: A Virtual Real Estate A(I)gent

Ata Çelen, Marc Pollefeys, Daniel Barath et al. (+1 more)

We introduce HouseTour, a method for spatially-aware 3D camera trajectory and natural language summary generation from a collection of images depicting an existing 3D space. Unlike existing vision-language models (VLMs), which struggle with geometric reasoning, our approach generates smooth video trajectories via a diffusion process constrained by known camera poses and integrates this information into the VLM for 3D-grounded descriptions. We synthesize the final video using 3D Gaussian splatting to render novel views along the trajectory. To support this task, we present the HouseTour dataset, which includes over 1,200 house-tour videos with camera poses, 3D reconstructions, and real estate descriptions. Experiments demonstrate that incorporating 3D camera trajectories into the text generation process improves performance over methods handling each task independently. We evaluate both individual and end-to-end performance, introducing a new joint metric. Our work enables automated, professional-quality video creation for real estate and touristic applications without requiring specialized expertise or equipment.
Read the full abstract

All authors Ata Çelen, Marc Pollefeys, Daniel Barath, Iro Armeni

Computation and Language

Language Models as Semantic Augmenters for Sequential Recommenders

Mahsa Valizadeh, Xiangjue Dong, Rui Tuo et al. (+1 more)

Large Language Models (LLMs) excel at capturing latent semantics and contextual relationships across diverse modalities. However, in modeling user behavior from sequential interaction data, performance often suffers when such semantic context is limited or absent. We introduce LaMAR, a LLM-driven semantic enrichment framework designed to enrich such sequences automatically. LaMAR leverages LLMs in a few-shot setting to generate auxiliary contextual signals by inferring latent semantic aspects of a user's intent and item relationships from existing metadata. These generated signals, such as inferred usage scenarios, item intents, or thematic summaries, augment the original sequences with greater contextual depth. We demonstrate the utility of this generated resource by integrating it into benchmark sequential modeling tasks, where it consistently improves performance. Further analysis shows that LLM-generated signals exhibit high semantic novelty and diversity, enhancing the representational capacity of the downstream models. This work represents a new data-centric paradigm where LLMs serve as intelligent context generators, contributing a new method for the semi-automatic creation of training data and language resources.
Read the full abstract

All authors Mahsa Valizadeh, Xiangjue Dong, Rui Tuo, James Caverlee

Computation and Language · Cross-listed from cs.AI

Subject-Event Ontology Without Global Time: Foundations and Execution Semantics

Alexander Boldachev

A formalization of a subject-event ontology is proposed for modeling complex dynamic systems without reliance on global time. Key principles: (1) event as an act of fixation - a subject discerns and fixes changes according to models (conceptual templates) available to them; (2) causal order via happens-before - the order of events is defined by explicit dependencies, not timestamps; (3) making the ontology executable via a declarative dataflow mechanism, ensuring determinism; (4) models as epistemic filters - a subject can only fix what falls under its known concepts and properties; (5) presumption of truth - the declarative content of an event is available for computation from the moment of fixation, without external verification. The formalization includes nine axioms (A1-A9), ensuring the correctness of executable ontologies: monotonicity of history (I1), acyclicity of causality (I2), traceability (I3). Special attention is given to the model-based approach (A9): event validation via schemas, actor authorization, automatic construction of causal chains (W3) without global time. Practical applicability is demonstrated on the boldsea system - a workflow engine for executable ontologies, where the theoretical constructs are implemented in BSL (Boldsea Semantic Language). The formalization is applicable to distributed systems, microservice architectures, DLT platforms, and multiperspectivity scenarios (conflicting facts from different subjects).
Read the full abstract

Computation and Language

From Local to Global: Revisiting Structured Pruning Paradigms for Large Language Models

Ziyan Wang, Enmao Diao, Qi Le et al. (+6 more)

Structured pruning is a practical approach to deploying large language models (LLMs) efficiently, as it yields compact, hardware-friendly architectures. However, the dominant local paradigm is task-agnostic: by optimizing layer-wise reconstruction rather than task objectives, it tends to preserve perplexity or generic zero-shot behavior but fails to capitalize on modest task-specific calibration signals, often yielding limited downstream gains. We revisit global structured pruning and present GISP, Global Iterative Structured Pruning, a post-training method that removes attention heads and MLP channels using first-order, loss-based important scores aggregated at the structure level with block-wise normalization. Built on this global importance metric, GISP adopts an iterative schedule, rather than one-shot pruning, stabilizes accuracy at higher sparsity, and mitigates perplexity collapse without requiring intermediate fine-tuning. Importantly, the iterative pruning forms nested subnetworks that support a ”prune-once, deploy-many” workflow. Furthermore, GISP defines structural importance directly with respect to a target loss, making it easy to adapt pruning to task-specific objectives. In this work, we use perplexity for language modeling and a margin-based objective for decision-style tasks. Extensive experiments show that across Llama2-7B/13B, Llama3-8B, and Mistral-0.3-7B, GISP consistently lowers WikiText-2 perplexity and improves on downstream accuracy, with especially strong gains at 40-50% sparsity; on DeepSeek-R1-Distill-Llama-3-8B and Qwen3-8B with GSM8K, task-aligned calibration substantially boosts exact-match accuracy. The implementation is available at https://github.com/uncc-efficient-ai/GISP.
Read the full abstract

All authors Ziyan Wang, Enmao Diao, Qi Le, Pu Wang, Minwoo Lee, Shu-ping Yeh, Evgeny Stupachenko, Hao Feng, Li Yang

Computation and Language

Is Multilingual LLM Watermarking Truly Multilingual? Scaling Robustness to 100+ Languages via Back-Translation

Asim Mohamed and Martin Gubri

Multilingual watermarking aims to make large language model (LLM) outputs traceable across languages, yet current methods still fall short. Despite claims of cross-lingual robustness, they are evaluated only on high-resource languages. We show that existing multilingual watermarking methods are not truly multilingual: they fail to remain robust under translation attacks in medium- and low-resource languages. We trace this failure to semantic clustering, which fails when the tokenizer vocabulary contains too few full-word tokens for a given language. To address this, we introduce STEAM, a detection method that uses Bayesian optimisation to search among 126 candidate languages for the back-translation that best recovers the watermark strength. It is compatible with any watermarking method, robust across different tokenizers and languages, non-invasive, and easily extendable to new languages. With average gains of +0.23 AUC and +37%p TPR@1%, STEAM provides a scalable approach toward fairer watermarking across the diversity of languages.
Read the full abstract

Computation and Language

SimBA: Simplifying Benchmark Analysis Using Performance Matrices Alone

Nishant Subramani, Alfredo Gomez and Mona Diab

Modern language models are evaluated on large benchmarks, which are difficult to make sense of, especially for model selection. Looking at the raw evaluation numbers themselves using a model-centric lens, we propose SimBA, a three phase framework to Simplify Benchmark Analysis. The three phases of SimBA are: stalk, where we conduct dataset & model comparisons, prowl, where we discover a representative subset, and pounce, where we use the representative subset to predict performance on a held-out set of models. Applying SimBA to three popular LM benchmarks: HELM, MMLU, and BigBenchLite reveals that across all three benchmarks, datasets and models relate strongly to one another (stalk). We develop an representative set discovery algorithm which covers a benchmark using raw evaluation scores alone. Using our algorithm, we find that with 6.25% (1/16), 1.7% (1/58), and 28.4% (21/74) of the datasets for HELM, MMLU, and BigBenchLite respectively, we achieve coverage levels of at least 95% (prowl). Additionally, using just these representative subsets, we can both preserve model ranks and predict performance on a held-out set of models with near zero mean-squared error (pounce). Taken together, SimBA can help model developers improve efficiency during model training and dataset creators validate whether their newly created dataset differs from existing datasets in a benchmark. Our code is open source, available at https://github.com/nishantsubramani/simba.
Read the full abstract

Computation and Language · Cross-listed from cs.CR

PLAGUE: Plug-and-play framework for Lifelong Adaptive Generation of Multi-turn Exploits

Neeladri Bhuiya, Madhav Aggarwal and Diptanshu Purwar

Large Language Models (LLMs) are improving at an exceptional rate. With the advent of agentic workflows, multi-turn dialogue has become the de facto mode of interaction with LLMs for completing long and complex tasks. While LLM capabilities continue to improve, they remain increasingly susceptible to jailbreaking, especially in multi-turn scenarios where harmful intent can be subtly injected across the conversation to produce nefarious outcomes. While single-turn attacks have been extensively explored, adaptability, efficiency and effectiveness continue to remain key challenges for their multi-turn counterparts. To address these gaps, we present PLAGUE, a novel plug-and-play framework for designing multi-turn attacks inspired by lifelong-learning agents. PLAGUE dissects the lifetime of a multi-turn attack into three carefully designed phases (Primer, Planner and Finisher) that enable a systematic and information-rich exploration of the multi-turn attack family. Evaluations show that red-teaming agents designed using PLAGUE achieve state-of-the-art jailbreaking results, improving attack success rates (ASR) by more than 30% across leading models in a lesser or comparable query budget. Particularly, PLAGUE enables an ASR (based on StrongReject) of 81.4% on OpenAI's o3 and 67.3% on Claude's Opus 4.1, two models that are considered highly resistant to jailbreaks in safety literature. Our work offers tools and insights to understand the importance of plan initialization, context optimization and lifelong learning in crafting multi-turn attacks for a comprehensive model vulnerability evaluation.
Read the full abstract

Computation and Language

Believe It or Not: How Deeply do LLMs Believe Implanted Facts?

Stewart Slocum, Julian Minder, Clément Dumas et al. (+4 more)

Knowledge editing techniques promise to implant new factual knowledge into large language models (LLMs). But do LLMs really believe these facts? We develop a framework to measure belief depth and use it to evaluate the success of knowledge editing techniques. We operationalize belief depth as the extent to which implanted knowledge 1) generalizes to related contexts (e.g. Fermi estimates several logical steps removed), 2) is robust to self-scrutiny and direct challenge, and 3) is represented similarly to genuine knowledge (as measured by linear probes). Our evaluations show that simple prompting and mechanistic editing techniques fail to implant knowledge deeply. In contrast, Synthetic Document Finetuning (SDF) - where models are trained on LLM-generated documents consistent with a fact - often succeeds at implanting beliefs that behave similarly to genuine knowledge. However, SDF's success is not universal, as implanted beliefs that contradict basic world knowledge are brittle and representationally distinct from genuine knowledge. Overall, our work introduces measurable criteria for belief depth and enables the rigorous evaluation necessary for deploying knowledge editing in real-world applications.
Read the full abstract

All authors Stewart Slocum, Julian Minder, Clément Dumas, Henry Sleight, Ryan Greenblatt, Samuel Marks, Rowan Wang

Computation and Language

AtlasKV: Augmenting LLMs with Billion-Scale Knowledge Graphs in 20GB VRAM

Haoyu Huang, Hong Ting Tsang, Jiaxin Bai et al. (+3 more)

Retrieval-augmented generation (RAG) has shown some success in augmenting large language models (LLMs) with external knowledge. However, as a non-parametric knowledge integration paradigm for LLMs, RAG methods heavily rely on external retrieval modules and the retrieved textual context prior. Especially for very large scale knowledge augmentation, they would introduce substantial inference latency due to expensive searches and much longer relevant context. In this paper, we propose a parametric knowledge integration method, called AtlasKV, a scalable, effective, and general way to augment LLMs with billion-scale knowledge graphs (KGs) (e.g. 1B triples) using very little GPU memory cost (e.g. less than 20GB VRAM). In AtlasKV, we introduce KG2KV and HiKVP to integrate KG triples into LLMs at scale with sub-linear time and memory complexity. It maintains strong knowledge grounding and generalization performance using the LLMs' inherent attention mechanism, and requires no external retrievers, long context priors, or retraining when adapting to new knowledge.
Read the full abstract

All authors Haoyu Huang, Hong Ting Tsang, Jiaxin Bai, Xi Peng, Gong Zhang, Yangqiu Song

Computation and Language

Diagnosing Representation Dynamics in NER Model Extension

Xirui Zhang, Philippe de La Chevasnerie and Benoit Fabre

Extending Named Entity Recognition (NER) models to new PII entities in noisy spoken-language data is a common need. We find that jointly fine-tuning a BERT model on standard semantic entities (PER, LOC, ORG) and new pattern-based PII (EMAIL, PHONE) results in minimal degradation for original classes. We investigate this "peaceful coexistence," hypothesizing that the model uses independent semantic vs. morphological feature mechanisms. Using an incremental learning setup as a diagnostic tool, we measure semantic drift and find two key insights. First, the LOC (location) entity is uniquely vulnerable due to a representation overlap with new PII, as it shares pattern-like features (e.g., postal codes). Second, we identify a "reverse O-tag representation drift." The model, initially trained to map PII patterns to 'O', blocks new learning. This is resolved only by unfreezing the 'O' tag's classifier, allowing the background class to adapt and "release" these patterns. This work provides a mechanistic diagnosis of NER model adaptation, highlighting feature independence, representation overlap, and 'O' tag plasticity. Work done based on data gathered by https://www.papernest.com
Read the full abstract

Computation and Language

Efficient Toxicity Detection in Gaming Chats: A Comparative Study of Embeddings, Fine-Tuned Transformers and LLMs

Yehor Tereshchenko and Mika Hämäläinen

This paper presents a comprehensive comparative analysis of Natural Language Processing (NLP) methods for automated toxicity detection in online gaming chats. Traditional machine learning models with embeddings, large language models (LLMs) with zero-shot and few-shot prompting, fine-tuned transformer models, and retrieval-augmented generation (RAG) approaches are evaluated. The evaluation framework assesses three critical dimensions: classification accuracy, processing speed, and computational costs. A hybrid moderation system architecture is proposed that optimizes human moderator workload through automated detection and incorporates continuous learning mechanisms. The experimental results demonstrate significant performance variations across methods, with fine-tuned DistilBERT achieving optimal accuracy-cost trade-offs. The findings provide empirical evidence for deploying cost-effective, efficient content moderation systems in dynamic online gaming environments.
Read the full abstract

Computation and Language

Select-Then-Decompose: From Empirical Analysis to Adaptive Selection Strategy for Task Decomposition in Large Language Models

Shuodi Liu, Yingzhuo Liu, Zi Wang et al. (+4 more)

Large language models (LLMs) have demonstrated remarkable reasoning and planning capabilities, driving extensive research into task decomposition. Existing task decomposition methods focus primarily on memory, tool usage, and feedback mechanisms, achieving notable success in specific domains, but they often overlook the trade-off between performance and cost. In this study, we first conduct a comprehensive investigation on task decomposition, identifying six categorization schemes. Then, we perform an empirical analysis of three factors that influence the performance and cost of task decomposition: categories of approaches, characteristics of tasks, and configuration of decomposition and execution models, uncovering three critical insights and summarizing a set of practical principles. Building on this analysis, we propose the Select-Then-Decompose strategy, which establishes a closed-loop problem-solving process composed of three stages: selection, execution, and verification. This strategy dynamically selects the most suitable decomposition approach based on task characteristics and enhances the reliability of the results through a verification module. Comprehensive evaluations across multiple benchmarks show that the Select-Then-Decompose consistently lies on the Pareto frontier, demonstrating an optimal balance between performance and cost. Our code is publicly available at https://github.com/summervvind/Select-Then-Decompose.
Read the full abstract

All authors Shuodi Liu, Yingzhuo Liu, Zi Wang, Yusheng Wang, Huijia Wu, Liuyu Xiang, Zhaofeng He

Computation and Language

CLAWS:Creativity detection for LLM-generated solutions using Attention Window of Sections

Keuntae Kim, Eunhye Jeong, Sehyeon Lee et al. (+2 more)

Recent advances in enhancing the reasoning ability of large language models (LLMs) have been remarkably successful. LLMs trained with reinforcement learning (RL) for reasoning demonstrate strong performance in challenging tasks such as mathematics and coding, even with relatively small model sizes. However, despite these improvements in task accuracy, the assessment of creativity in LLM generations has been largely overlooked in reasoning tasks, in contrast to writing tasks. The lack of research on creativity assessment in reasoning primarily stems from two challenges: (1) the difficulty of defining the range of creativity, and (2) the necessity of human evaluation in the assessment process. To address these challenges, we propose CLAWS, a method that defines and classifies mathematical solutions into typical, creative, and hallucinated categories without human evaluation, by leveraging attention weights across prompt sections and output. CLAWS outperforms five existing white-box detection methods (Perplexity, Logit Entropy, Window Entropy, Hidden Score, and Attention Score) on five 7-8B math RL models (DeepSeek, Qwen, Mathstral, OpenMath2, and Oreal). We validate CLAWS on 4545 math problems collected from 181 math contests (AJHSME, AMC, AIME).
Read the full abstract

All authors Keuntae Kim, Eunhye Jeong, Sehyeon Lee, Seohee Yoon, Yong Suk Choi

Computation and Language

JT-Safe: Intrinsically Enhancing the Safety and Trustworthiness of LLMs

Junlan Feng, Fanyu Meng, Chong Long et al. (+12 more)

The hallucination and credibility concerns of large language models (LLMs) are global challenges that the industry is collectively addressing. Recently, a significant amount of advances have been made on post-training and inference techniques to mitigate these challenges. However, it is widely agreed that unsafe and hallucinations of LLMs intrinsically originate from pre-training, involving pre-training data and the next-token prediction learning mechanism. In this paper, we focus on enhancing pre-training data to improve the trustworthiness and safety of LLMs. Since the data is vast, it's almost impossible to entirely purge the data of factual errors, logical inconsistencies, or distributional biases. Moreover, the pre-training data lack grounding in real-world knowledge. Each piece of data is treated as a sequence of tokens rather than as a representation of a part of the world. To overcome these issues, we propose approaches to enhancing our pre-training data with its context in the world and increasing a substantial amount of data reflecting industrial scenarios. We argue that most source data are created by the authors for specific purposes in a certain spatial-temporal context. They have played a role in the real world. By incorporating related world context information, we aim to better anchor pre-training data within real-world scenarios, thereby reducing uncertainty in model training and enhancing the model's safety and trustworthiness. We refer to our Data with World Context as DWC. We continue pre-training an earlier checkpoint of JT-35B-Base with 1.5 trillion of DWC tokens. We introduce our post-training procedures to activate the potentials of DWC. Compared with the Qwen model of a similar scale, JT-Safe-35B achieves an average performance improvement of 1.79% on the Safety and Trustworthy evaluation benchmarks, while being pretrained with only 6.2 trillion tokens.
Read the full abstract

All authors Junlan Feng, Fanyu Meng, Chong Long, Pengyu Cong, Duqing Wang, Yan Zheng, Yuyao Zhang, Xuanchang Gao, Ye Yuan, Yunfei Ma, Zhijie Ren, Fan Yang, Na Wu, Di Jin, Chao Deng

Computation and Language · Cross-listed from cs.CV

Glyph: Scaling Context Windows via Visual-Text Compression

Jiale Cheng, Yusen Liu, Xinyu Zhang et al. (+11 more)

Large language models (LLMs) increasingly rely on long-context modeling for tasks such as document understanding, code analysis, and multi-step reasoning. However, scaling context windows to the million-token level brings prohibitive computational and memory costs, limiting the practicality of long-context LLMs. In this work, we take a different perspective-visual context scaling-to tackle this challenge. Instead of extending token-based sequences, we propose Glyph, a framework that renders long texts into images and processes them with vision-language models (VLMs). This approach substantially compresses textual input while preserving semantic information, and we further design an LLM-driven genetic search to identify optimal visual rendering configurations for balancing accuracy and compression. Through extensive experiments, we demonstrate that our method achieves 3-4x token compression while maintaining accuracy comparable to leading LLMs such as Qwen3-8B on various long-context benchmarks. This compression also leads to around 4x faster prefilling and decoding, and approximately 2x faster SFT training. Furthermore, under extreme compression, a 128K-context VLM could scale to handle 1M-token-level text tasks. In addition, the rendered text data benefits real-world multimodal tasks, such as document understanding. Our code and model are released at https://github.com/thu-coai/Glyph.
Read the full abstract

All authors Jiale Cheng, Yusen Liu, Xinyu Zhang, Yulin Fei, Wenyi Hong, Ruiliang Lyu, Weihan Wang, Zhe Su, Xiaotao Gu, Xiao Liu, Yushi Bai, Jie Tang, Hongning Wang, Minlie Huang

Computation and Language

Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics

Akshara Prabhakar, Roshan Ram, Zixiang Chen et al. (+5 more)

As information grows exponentially, enterprises face increasing pressure to transform unstructured data into coherent, actionable insights. While autonomous agents show promise, they often struggle with domain-specific nuances, intent alignment, and enterprise integration. We present Enterprise Deep Research (EDR), a multi-agent system that integrates (1) a Master Planning Agent for adaptive query decomposition, (2) four specialized search agents (General, Academic, GitHub, LinkedIn), (3) an extensible MCP-based tool ecosystem supporting NL2SQL, file analysis, and enterprise workflows, (4) a Visualization Agent for data-driven insights, and (5) a reflection mechanism that detects knowledge gaps and updates research direction with optional human-in-the-loop steering guidance. These components enable automated report generation, real-time streaming, and seamless enterprise deployment, as validated on internal datasets. On open-ended benchmarks including DeepResearch Bench and DeepConsult, EDR outperforms state-of-the-art agentic systems without any human steering. We release the EDR framework and benchmark trajectories to advance research on multi-agent reasoning applications. Code at https://github.com/SalesforceAIResearch/enterprise-deep-research and Dataset at https://huggingface.co/datasets/Salesforce/EDR-200
Read the full abstract

All authors Akshara Prabhakar, Roshan Ram, Zixiang Chen, Silvio Savarese, Frank Wang, Caiming Xiong, Huan Wang, Weiran Yao

Computation and Language

What Makes AI Research Replicable? Executable Knowledge Graphs as Scientific Knowledge Representations

Yujie Luo, Zhuoyun Yu, Xuehai Wang et al. (+6 more)

Replicating AI research is a crucial yet challenging task for large language model (LLM) agents. Existing approaches often struggle to generate executable code, primarily due to insufficient background knowledge and the limitations of retrieval-augmented generation (RAG) methods, which fail to capture latent technical details hidden in referenced papers. Furthermore, previous approaches tend to overlook valuable implementation-level code signals and lack structured knowledge representations that support multi-granular retrieval and reuse. To overcome these challenges, we propose Executable Knowledge Graphs (xKG), a pluggable, paper-centric knowledge base that automatically integrates code snippets and technical insights extracted from scientific literature. When integrated into three agent frameworks with two different LLMs, xKG shows substantial performance gains (10.9% with o3-mini) on PaperBench, demonstrating its effectiveness as a general and extensible solution for automated AI research replication. Code is available at https://github.com/zjunlp/xKG.
Read the full abstract

All authors Yujie Luo, Zhuoyun Yu, Xuehai Wang, Yuqi Zhu, Ningyu Zhang, Lanning Wei, Lun Du, Da Zheng, Huajun Chen

Computation and Language

Foundational Automatic Evaluators: Scaling Multi-Task Generative Evaluator Training for Reasoning-Centric Domains

Austin Xu, Xuan-Phi Nguyen, Yilun Zhou et al. (+3 more)

Finetuning specialized generative evaluators has emerged as a popular paradigm to meet the increasing demand for scalable evaluation during both training and test-time. However, recent work has largely focused on applying new methodology, such as reinforcement learning (RL), to training evaluators, shying away from large-scale, data-driven development. In this work, we focus on data scaling, curating a set of 2.5M samples spanning five unique evaluation tasks (pairwise, step-level, reference-free and reference-based verification, and single rating) and multiple domains focused on reasoning evaluation. With our data, we train Foundational Automatic Reasoning Evaluators (FARE), a family of 8B and 20B (with 3.6B active) parameter evaluators, with a simple iterative rejection-sampling supervised finetuning (SFT) approach. FARE-8B challenges larger specialized RL-trained evaluators and FARE-20B sets the new standard for open-source evaluators, surpassing specialized 70B+ evaluators. Beyond static benchmarks, we evaluate FARE in real-world tasks: As inference-time rerankers, FARE-20B achieves near-oracle performance on MATH. As verifiers in RL training, FARE improves the downstream RL-trained model performance by up to 14.1% vs. string-matching verifiers. When initialized from FARE, a continually-finetuned FARE-Code outperforms gpt-oss-20B by 65% on evaluating test-case quality.
Read the full abstract

All authors Austin Xu, Xuan-Phi Nguyen, Yilun Zhou, Chien-Sheng Wu, Caiming Xiong, Shafiq Joty

Computation and Language · Cross-listed from cs.CV

UltraCUA: A Foundation Model for Computer Use Agents with Hybrid Action

Yuhao Yang, Zhen Yang, Zi-Yi Dou et al. (+10 more)

Computer-use agents face a fundamental limitation. They rely exclusively on primitive GUI actions (click, type, scroll), creating brittle execution chains prone to cascading failures. While API-driven agents harness rich capabilities through structured interfaces and tools, computer-use agents remain constrained to low-level visual interactions. We present UltraCUA, a foundation model that transcends this limitation through hybrid action-seamlessly unifying primitive GUI operations with high-level tool execution. Our innovation rests on four critical advances. First, an automated pipeline extracts and scales tool capabilities from software documentation and code repositories. Second, a synthetic data engine produces 17,000+ verifiable tasks capturing real-world computer-use complexity. Third, comprehensive hybrid action trajectory collection incorporates both GUI primitives and strategic tool calls. Fourth, a two-stage training methodology combines supervised fine-tuning with online reinforcement learning, enabling intelligent action selection between GUI and API. Evaluation with our 7B and 32B UltraCUA models reveals transformative performance gains. On OSWorld, UltraCUA achieves 22% relative improvement while executing 11% faster than existing approaches, averagely. Cross-domain validation on WindowsAgentArena demonstrates robust generalization with 21.7% success rate, surpassing Windows-trained baselines. The hybrid action paradigm proves essential, reducing error propagation while improving execution efficiency. This work establishes a scalable paradigm bridging primitive GUI interactions and high-level tool intelligence, enabling more resilient and adaptable computer use agents for diverse environments and complex real-world tasks.
Read the full abstract

All authors Yuhao Yang, Zhen Yang, Zi-Yi Dou, Anh Nguyen, Keen You, Omar Attia, Andrew Szot, Michael Feng, Ram Ramrakhya, Alexander Toshev, Chao Huang, Yinfei Yang, Zhe Gan

Computation and Language · Cross-listed from cs.LG

Mapping Post-Training Forgetting in Language Models at Scale

Jackson Harmon, Andreas Hochlehnert, Matthias Bethge et al. (+1 more)

Scaled post-training now drives many of the largest capability gains in language models (LMs), yet its effect on pretrained knowledge remains poorly understood. Not all forgetting is equal: Forgetting one fact (e.g., a U.S. president or an API call) does not "average out" by recalling another. Hence, we propose a sample-wise paradigm to measure what is forgotten and when backward transfer occurs. Our metric counts 1->0 transitions (correct before post-training, incorrect after) to quantify forgetting and 0->1 transitions to quantify backward transfer. Traditional task averages conflate these effects and obscure large changes. For multiple-choice benchmarks, we add chance-adjusted variants that subtract the expected contribution of random guessing from pre- and post-training accuracies. We apply this framework across post-training stages, model sizes, and data scales. Our large-scale analysis shows that: (1) Domain-continual pretraining induces moderate forgetting with low-to-moderate backward transfer; (2) RL/SFT post-training applied to base models and Instruction tuning yields moderate-to-large backward transfer on math and logic with overall low-to-moderate forgetting; (3) Applying RL/SFT to instruction-tuned models is sensitive on data scale: at small scales, both forgetting and backward transfer are small; at larger scales, effects are mixed and warrant further study with better controls; (4) Model merging does not reliably mitigate forgetting. Overall, our framework offers a practical yardstick for mapping how post-training alters pretrained knowledge at scale – enabling progress towards generally capable AI systems.
Read the full abstract

All authors Jackson Harmon, Andreas Hochlehnert, Matthias Bethge, Ameya Prabhu

Computation and Language

Evaluating Medical LLMs by Levels of Autonomy: A Survey Moving from Benchmarks to Applications

Xiao Ye, Jacob Dineen, Zhaonan Li et al. (+11 more)

Medical Large language models achieve strong scores on standard benchmarks; however, the transfer of those results to safe and reliable performance in clinical workflows remains a challenge. This survey reframes evaluation through a levels-of-autonomy lens (L0-L3), spanning informational tools, information transformation and aggregation, decision support, and supervised agents. We align existing benchmarks and metrics with the actions permitted at each level and their associated risks, making the evaluation targets explicit. This motivates a level-conditioned blueprint for selecting metrics, assembling evidence, and reporting claims, alongside directions that link evaluation to oversight. By centering autonomy, the survey moves the field beyond score-based claims toward credible, risk-aware evidence for real clinical use.
Read the full abstract

All authors Xiao Ye, Jacob Dineen, Zhaonan Li, Zhikun Xu, Weiyu Chen, Shijie Lu, Yuxi Huang, Ming Shen, Phu Tran, Ji-Eun Irene Yum, Muhammad Ali Khan, Muhammad Umar Afzal, Irbaz Bin Riaz, Ben Zhou

Computation and Language · Cross-listed from cs.CR

VERA-V: Variational Inference Framework for Jailbreaking Vision-Language Models

Qilin Liao, Anamika Lochab and Ruqi Zhang

Vision-Language Models (VLMs) extend large language models with visual reasoning, but their multimodal design also introduces new, underexplored vulnerabilities. Existing multimodal red-teaming methods largely rely on brittle templates, focus on single-attack settings, and expose only a narrow subset of vulnerabilities. To address these limitations, we introduce VERA-V, a variational inference framework that recasts multimodal jailbreak discovery as learning a joint posterior distribution over paired text-image prompts. This probabilistic view enables the generation of stealthy, coupled adversarial inputs that bypass model guardrails. We train a lightweight attacker to approximate the posterior, allowing efficient sampling of diverse jailbreaks and providing distributional insights into vulnerabilities. VERA-V further integrates three complementary strategies: (i) typography-based text prompts that embed harmful cues, (ii) diffusion-based image synthesis that introduces adversarial signals, and (iii) structured distractors to fragment VLM attention. Experiments on HarmBench and HADES benchmarks show that VERA-V consistently outperforms state-of-the-art baselines on both open-source and frontier VLMs, achieving up to 53.75% higher attack success rate (ASR) over the best baseline on GPT-4o. We include the code on the project page available here: https://github.com/kxwhiowo/VERA-V
Read the full abstract

Computation and Language

Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations

Tong Chen, Akari Asai, Luke Zettlemoyer et al. (+2 more)

Language models often generate factually incorrect information unsupported by their training data, a phenomenon known as extrinsic hallucination. Existing mitigation approaches often degrade performance on open-ended generation and downstream tasks, limiting their practical utility. We propose an online reinforcement learning method using a novel binary retrieval-augmented reward (RAR) to address this tradeoff. Unlike continuous reward schemes, our approach assigns a reward of one only when the model's output is entirely factually correct, and zero otherwise. We evaluate our method on Qwen3 reasoning models across diverse tasks. For open-ended generation, binary RAR achieves a 39.3% reduction in hallucination rates, substantially outperforming both supervised training and continuous-reward RL baselines. In short-form question answering, the model learns calibrated abstention, strategically outputting "I don't know" when faced with insufficient parametric knowledge. This yields 44.4% and 21.7% fewer incorrect answers on PopQA and GPQA, respectively. Crucially, these factuality gains come without performance degradation on instruction following, math, or code, whereas continuous-reward RL, despite improving factuality, induces quality regressions.
Read the full abstract

All authors Tong Chen, Akari Asai, Luke Zettlemoyer, Hannaneh Hajishirzi, Faeze Brahman

Computation and Language

AcademicEval: Live Long-Context LLM Benchmark

Haozhen Zhang, Tao Feng, Pengrui Han et al. (+1 more)

Large Language Models (LLMs) have recently achieved remarkable performance in long-context understanding. However, current long-context LLM benchmarks are limited by rigid context length, labor-intensive annotation, and the pressing challenge of label leakage issues during LLM training. Therefore, we propose AcademicEval, a live benchmark for evaluating LLMs over long-context generation tasks. AcademicEval adopts papers on arXiv to introduce several academic writing tasks with long-context inputs, i.e., Title, Abstract, Introduction, and Related Work, which cover a wide range of abstraction levels and require no manual labeling. Moreover, AcademicEval integrates high-quality and expert-curated few-shot demonstrations from a collected co-author graph to enable flexible context length. Especially, AcademicEval features an efficient live evaluation, ensuring no label leakage. We conduct a holistic evaluation on AcademicEval, and the results illustrate that LLMs perform poorly on tasks with hierarchical abstraction levels and tend to struggle with long few-shot demonstrations, highlighting the challenge of our benchmark. Through experimental analysis, we also reveal some insights for enhancing LLMs' long-context modeling capabilities. Code is available at https://github.com/ulab-uiuc/AcademicEval
Read the full abstract

All authors Haozhen Zhang, Tao Feng, Pengrui Han, Jiaxuan You

Computation and Language

PANER: A Paraphrase-Augmented Framework for Low-Resource Named Entity Recognition

Nanda Kumar Rengarajan, Jun Yan and Chun Wang

Named Entity Recognition (NER) is a critical task that requires substantial annotated data, making it challenging in low-resource scenarios where label acquisition is expensive. While zero-shot and instruction-tuned approaches have made progress, they often fail to generalize to domain-specific entities and do not effectively utilize limited available data. We present a lightweight few-shot NER framework that addresses these challenges through two key innovations: (1) a new instruction tuning template with a simplified output format that combines principles from prior IT approaches to leverage the large context window of recent state-of-the-art LLMs; (2) introducing a strategic data augmentation technique that preserves entity information while paraphrasing the surrounding context, thereby expanding our training data without compromising semantic relationships. Experiments on benchmark datasets show that our method achieves performance comparable to state-of-the-art models on few-shot and zero-shot tasks, with our few-shot approach attaining an average F1 score of 80.1 on the CrossNER datasets. Models trained with our paraphrasing approach show consistent improvements in F1 scores of up to 17 points over baseline versions, offering a promising solution for groups with limited NER training data and compute power.
Read the full abstract

Computation and Language

QueST: Incentivizing LLMs to Generate Difficult Problems

Hanxu Hu, Xingxing Zhang, Jannis Vamvas et al. (+2 more)

Large Language Models have achieved strong performance on reasoning tasks, solving competition-level coding and math problems. However, their scalability is limited by human-labeled datasets and the lack of large-scale, challenging coding problem training data. Existing competitive coding datasets contain only thousands to tens of thousands of problems. Previous synthetic data generation methods rely on either augmenting existing instruction datasets or selecting challenging problems from human-labeled data. In this paper, we propose QueST, a novel framework which combines difficulty-aware graph sampling and difficulty-aware rejection fine-tuning that directly optimizes specialized generators to create challenging coding problems. Our trained generators demonstrate superior capability compared to even GPT-4o at creating challenging problems that benefit downstream performance. We leverage QueST to generate large-scale synthetic coding problems, which we then use to distill from strong teacher models with long chain-of-thought or to conduct reinforcement learning for smaller models, proving effective in both scenarios. Our distillation experiments demonstrate significant performance gains. Specifically, after fine-tuning Qwen3-8B-base on 100K difficult problems generated by QueST, we surpass the performance of the original Qwen3-8B on LiveCodeBench. With an additional 112K examples (i.e., 28K human-written problems paired with multiple synthetic solutions), our 8B model matches the performance of the much larger DeepSeek-R1-671B. These findings indicate that generating complex problems via QueST offers an effective and scalable approach to advancing the frontiers of competitive coding and reasoning for large language models.
Read the full abstract

All authors Hanxu Hu, Xingxing Zhang, Jannis Vamvas, Rico Sennrich, Furu Wei

Computation and Language · Cross-listed from cs.AI

Contextual Attention Modulation: Towards Efficient Multi-Task Adaptation in Large Language Models

Dayan Pan, Zhaoyang Fu, Jingyuan Wang et al. (+3 more)

Large Language Models (LLMs) possess remarkable generalization capabilities but struggle with multi-task adaptation, particularly in balancing knowledge retention with task-specific specialization. Conventional fine-tuning methods suffer from catastrophic forgetting and substantial resource consumption, while existing parameter-efficient methods perform suboptimally in complex multi-task scenarios. To address this, we propose Contextual Attention Modulation (CAM), a novel mechanism that dynamically modulates the representations of self-attention modules in LLMs. CAM enhances task-specific features while preserving general knowledge, thereby facilitating more effective and efficient adaptation. For effective multi-task adaptation, CAM is integrated into our Hybrid Contextual Attention Modulation (HyCAM) framework, which combines a shared, full-parameter CAM module with multiple specialized, lightweight CAM modules, enhanced by a dynamic routing strategy for adaptive knowledge fusion. Extensive experiments on heterogeneous tasks, including question answering, code generation, and logical reasoning, demonstrate that our approach significantly outperforms existing approaches, achieving an average performance improvement of 3.65%. The implemented code and data are available to ease reproducibility at https://github.com/Applied-Machine-Learning-Lab/HyCAM.
Read the full abstract

All authors Dayan Pan, Zhaoyang Fu, Jingyuan Wang, Xiao Han, Yue Zhu, Xiangyu Zhao