Large language models (LLMs) are increasingly integrated with specialized external tools, yet many tasks demand zero-shot tool usage with minimal or noisy documentation. Existing solutions rely on manual rewriting or labeled data for validation, making them inapplicable in true zero-shot settings. To address these challenges, we propose PLAY2PROMPT, an automated framework that systematically “plays” with each tool to explore its input-output behaviors. Through this iterative trial-and-error process, PLAY2PROMPT refines tool documentation and generates usage examples without any labeled data. These examples not only guide LLM inference but also serve as validation to further enhance tool utilization. Extensive experiments on real-world tasks demonstrate that PLAY2PROMPT significantly improves zero-shot tool performance across both open and closed models, offering a scalable and effective solution for domain-specific tool integration.
论文检索
输入标题、作者或关键词，从 3,086 篇学术成果中精准定位
Speculative Decoding (SD) enforces strict distributional equivalence to the target model when accepting candidate tokens. While it maintains the target model’s generation quality, this strict equivalence limits the speedup achievable by SD and prevents users from trading deviations from the target distribution in exchange for further inference speed gains. To address these limitations, we introduce Fuzzy Speculative Decoding (FSD) - a decoding algorithm that generalizes SD by accepting candidate tokens based on the divergences between the target and draft model distributions. By allowing for controlled divergence from the target model, FSD enables users to flexibly trade generation quality for inference speed. Across several benchmarks, our method is able to achieve significant runtime improvements of over 5 tokens per second faster than SD at only an approximate 2% absolute reduction in benchmark accuracy. In many cases, FSD is even able to match SD benchmark accuracy at over 2 tokens per second faster, demonstrating that distributional equivalence is not necessary to maintain target model performance. Furthermore, FSD can be seamlessly integrated into existing SD extensions; we demonstrate this by applying FSD to EAGLE-2, greatly enhancing this existing extension’s efficiency while allowing it to leverage FSD’s tunable quality-speed trade-off.
M^{2}-TabFact: Multi-Document Multi-Modal Fact Verification with Visual and Textual Representations of Tabular Data
PDF ↗Tabular data is used to store information in many real-world systems ranging from finance to healthcare. However, such structured data is often communicated to humans in visually interpretable formats (e.g. charts and textual paragraphs), making it imperative that fact-checking models should be able to reason over multiple pieces of structured evidence presented across different modalities. In this paper, we propose Multi-Document Multi-Modal Table-based Fact Verification (M^{2}-TabFact), a challenging fact verification task that requires jointly reasoning over visual and textual representations of structured data. We design an automatic data generation pipeline that converts existing tabular data into descriptive visual and textual evidence. We then use Large Language Models to generate complex claims that depend on multi-document, multi-modal evidence. In total, we create 8,856 pairs of complex claims and multi-modal evidence through this procedure and systematically evaluate M^{2}-TabFact with a set of strong vision-language models (VLM). We find that existing VLMs have large gaps in fact verification performance compared to humans. Moreover, we find that they are imbalanced when it comes to their ability to handle reason about different modalities, and currently struggle to reason about information extracted from multiple documents.
In recent years, large language models (LLMs) have achieved breakthrough progress in many dialogue generation tasks. However, their lack of emotion and fine-grained role awareness limits the model’s ability to provide personalized and diverse interactions further. Current methods face high costs in collecting high-quality annotated data for scenarios such as role-playing, and traditional human alignment methods are difficult to deploy due to the inherent diversity of model behavior in role-playing scenarios. Inspired by the alignment of models for safety behaviors through RLHF (Reinforcement Learning from Human Feedback), in this paper, we revisit model role-playing behavior from the perspective of persona alignment and propose a novel annotation-free framework named Persona-Aware Contrastive Learning (PCL) to align LLMs’ behavior during role-playing, enhancing the model’s role consistency. Specifically, we first design a role chain method to encourage the model to self-question based on the role characteristics and dialogue context to adjust personality consistency. Then, we further enhance the model’s role-playing strategy through iterative adversarial modeling between the use of role characteristics and not. Experiments on both black-box and white-box LLMs show that LLMs equipped with PCL significantly outperform vanilla LLMs under automatic evaluation methods (CharEval & GPT-4) and human expert evaluation.
Protecting Users From Themselves: Safeguarding Contextual Privacy in Interactions with Conversational Agents
PDF ↗Conversational agents are increasingly woven into individuals’ personal lives, yet users often underestimate the privacy risks associated with them. The moment users share information with these agents —such as large language models (LLMs)— their private information becomes vulnerable to exposure. In this paper, we characterize the notion of contextual privacy for user interactions with LLM-based Conversational Agents (LCAs). It aims to minimize privacy risks by ensuring that users (sender) disclose only information that is both relevant and necessary for achieving their intended goals when interacting with LCAs (untrusted receivers). Through a formative design user study, we observe how even “privacy-conscious” users inadvertently reveal sensitive information through indirect disclosures. Based on insights from this study, we propose a locally deployable framework that operates between users and LCAs, identifying and reformulating out-of-context information in user prompts. Our evaluation using examples from ShareGPT shows that lightweight models can effectively implement this framework, achieving strong gains in contextual privacy while preserving the user’s intended interaction goals. Notably, about 76% of participants in our human evaluation preferred the reformulated prompts over the original ones, validating the usability and effectiveness of contextual privacy in our proposed framework. We open source the code at https://github.com/IBM/contextual-privacy-LLM.
Internal world models (WMs) enable agents to understand the world’s state and predict transitions, serving as the basis for advanced deliberative reasoning.Recent large Vision-Language Models (VLMs), such as GPT-4o and Gemini, exhibit potential as general-purpose WMs. While the latest studies have evaluated and shown limitations in specific capabilities such as visual understanding, a systematic evaluation of VLMs’ fundamental WM abilities remains absent. Drawing on comparative psychology and cognitive science, we propose a two-stage framework that assesses **perception** (visual, spatial, temporal, quantitative, and motion) and **prediction** (mechanistic simulation, transitive inference, compositional inference) to provide an atomic evaluation of VLMs as WMs. Guided by this framework, we introduce **WM-ABench**, a large-scale benchmark comprising 23 fine-grained evaluation dimensions across 6 diverse simulated environments with controlled counterfactual simulations. Through 660 experiments on 15 latest commercial and open-source VLMs, we find that these models exhibit striking limitations in basic world modeling abilities. For instance, all models perform at near-random accuracy when distinguishing motion trajectories. Additionally, they lack disentangled understanding—e.g., they tend to believe blue objects move faster than green ones. More rich results and analyses reveal significant gaps between VLMs and human-level world modeling.
Existing fairness benchmarks for large language models (LLMs) primarily focus on simple tasks, such as multiple-choice questions, overlooking biases that may arise in more complex scenarios like long-text generation. To address this gap, we introduce the Long Text Fairness Test (LTF-TEST), a framework that evaluates biases in LLMs through essay-style prompts. LTF-TEST covers 14 topics and 10 demographic axes, including gender and race, resulting in 11,948 samples. By assessing both model responses and the reasoning behind them, LTF-TEST uncovers subtle biases that are difficult to detect in simple responses. In our evaluation of five recent LLMs, including GPT-4o and LLaMA3, we identify two key patterns of bias. First, these models frequently favor certain demographic groups in their responses. Second, they show excessive sensitivity toward traditionally disadvantaged groups, often providing overly protective responses while neglecting others. To mitigate these biases, we propose REGARD-FT, a finetuning approach that pairs biased prompts with neutral responses. REGARD-FT reduces gender bias by 34.6% and improves performance by 1.4 percentage points on the BBQ benchmark, offering a promising approach to addressing biases in long-text generation tasks.
This study explores a novel approach to enhance the performance of Large Language Models (LLMs) through the optimization of input data within prompts. While previous research has primarily focused on refining instruction components and augmenting input data with in-context examples, our work investigates the potential benefits of optimizing the input data itself. We introduce a two-pronged strategy for input data optimization: content engineering and structural reformulation. Content engineering involves imputing missing values, removing irrelevant attributes, and enriching profiles by generating additional information inferred from existing attributes. Subsequent to content engineering, structural reformulation is applied to optimize the presentation of the modified content to LLMs, given their sensitivity to input format. Our findings suggest that these optimizations can significantly improve the performance of LLMs in various tasks, offering a promising avenue for future research in prompt engineering. The source code is available at https://github.com/glin2229/Automatic-Data-Optimization.
Large language models (LLMs) excel in many tasks but struggle to accurately quantify uncertainty in their generated responses. This limitation makes it challenging to detect misinformation and ensure reliable decision-making. Existing uncertainty quantification (UQ) methods for LLMs are primarily prompt-wise rather than response-wise, often requiring multiple response samples, which leads to inefficiency. Moreover, LLMs have been shown to be overconfident, particularly when using reasoning steps to derive their answers. In this work, we introduce a novel approach to quantify response-wise uncertainty by integrating LLMs’ inherent reasoning capabilities through Chain-of-Thought (CoT) into the UQ process. Our CoT-UQ framework captures critical information during inference by extracting keywords from each reasoning step and assessing their importance to the final answer. The uncertainty scores of keywords are then aggregated based on their significance to produce a final uncertainty estimate. We conduct extensive experiments based on Llama Family with model sizes varying from 8B to 13B across logical and mathematical reasoning tasks. Experimental results demonstrate that CoT-UQ significantly outperforms existing UQ methods, achieving an average improvement of 5.9% AUROC compared to current UQ methods.
This study investigates the halo effect in AI-driven hiring evaluations using Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs). Through experiments with hypothetical job applications, we examined how these models’ evaluations are influenced by non-job-related information, including extracurricular activities and social media images. By analyzing models’ responses to Likert-scale questions across different competency dimensions, we found that AI models exhibit significant halo effects, particularly in image-based evaluations, while text-based assessments showed more resistance to bias. The findings demonstrate that supplementary multimodal information can substantially influence AI hiring decisions, highlighting potential risks in AI-based recruitment systems.
Recently, there has been growing interest in leveraging large language models (LLMs) to generate symbolic world models from textual descriptions. Although LLMs have been extensively explored in the context of world modeling, prior studies encountered several challenges, including evaluation randomness, dependence on indirect metrics, and a limited domain scope. To address these limitations, we introduce a novel benchmark, Text2World, based on planning domain definition language (PDDL), featuring hundreds of diverse domains and employing multi-criteria, execution-based metrics for a more robust evaluation. We benchmark current LLMs using Text2World and find that reasoning models trained with large-scale reinforcement learning outperform others. However, even the best-performing model still demonstrates limited capabilities in world modeling. Building on these insights, we examine several promising strategies to enhance the world modeling capabilities of LLMs, including test-time scaling, agent training, and more. We hope that Text2World can serve as a crucial resource, laying the groundwork for future research in leveraging LLMs as world models.
BridG MT: Enhancing LLMs’ Machine Translation Capabilities with Sentence Bridging and Gradual MT
PDF ↗Recent Large Language Models (LLMs) have demonstrated impressive translation performance without requiring fine-tuning on additional parallel corpora. However, they still face significant challenges in certain scenarios, particularly when translating low-resource languages. A common approach to address this issue is to provide external knowledge, such as few-shot examples, to assist LLMs in translating specific source sentences. However, this method is fundamentally limited by the quality or quantity of relevant sources, which cannot always be guaranteed. To reduce LLMs’ reliance on external sources, we propose BridG MT, a method that combines Sentence Bridging, which generates a sequence of sentences as a bridge that gradually transition from easy-to-translate to more difficult, and Gradual MT, which sequentially translates these sentences using earlier translations as few-shot examples for subsequent ones. Experiments conducted on four LLMs across seven languages demonstrate that our method effectively enhances translation performance, even outperforming translation methods that rely on a large number of few-shot examples.
The Million Authors Corpus: A Cross-Lingual and Cross-Domain Wikipedia Dataset for Authorship Verification
PDF ↗Authorship verification (AV) is a crucial task for applications like identity verification, plagiarism detection, and AI-generated text identification. However, datasets for training and evaluating AV models are primarily in English and primarily in a single domain. This precludes analysis of AV techniques for generalizability and can cause seemingly valid AV solutions to, in fact, rely on topic-based features rather than actual authorship features. To address this limitation, we introduce the Million Authors Corpus (), a novel dataset encompassing contributions from dozens of languages on Wikipedia. It includes only long and contiguous textual chunks taken from Wikipedia edits and links those texts to their authors. includes 60.08M textual chunks, contributed by 1.29M Wikipedia authors. It enables broad-scale cross-lingual and cross-domain AV evaluation to ensure accurate analysis of model capabilities that are not overly optimistic. We provide baseline evaluations using state-of-the-art AV models as well as information retrieval models that are not AV-specific in order to demonstrate ‘s unique cross-lingual and cross-domain ablation capabilities.
When Should Dense Retrievers Be Updated in Evolving Corpora? Detecting Out-of-Distribution Corpora Using GradNormIR
PDF ↗Dense retrievers encode texts into embeddings to efficiently retrieve relevant documents from large databases in response to user queries. However, real-world corpora continually evolve, leading to a shift from the original training distribution of the retriever. Without timely updates or retraining, indexing newly emerging documents can degrade retrieval performance for future queries. Thus, identifying when a dense retriever requires an update is critical for maintaining robust retrieval systems. In this paper, we propose a novel task of predicting whether a corpus is out-of-distribution (OOD) relative to a dense retriever before indexing. Addressing this task allows us to proactively manage retriever updates, preventing potential retrieval failures. We introduce GradNormIR, an unsupervised approach that leverages gradient norms to detect OOD corpora effectively. Experiments on the BEIR benchmark demonstrate that GradNormIR enables timely updates of dense retrievers in evolving document collections, significantly enhancing retrieval robustness and efficiency.
Dictionaries to the Rescue: Cross-Lingual Vocabulary Transfer for Low-Resource Languages Using Bilingual Dictionaries
PDF ↗Cross-lingual vocabulary transfer plays a promising role in adapting pre-trained language models to new languages, including low-resource languages.Existing approaches that utilize monolingual or parallel corpora face challenges when applied to languages with limited resources.In this work, we propose a simple yet effective vocabulary transfer method that utilizes bilingual dictionaries, which are available for many languages, thanks to descriptive linguists.Our proposed method leverages a property of BPE tokenizers where removing a subword from the vocabulary causes a fallback to shorter subwords.The embeddings of target subwords are estimated iteratively by progressively removing them from the tokenizer.The experimental results show that our approach outperforms existing methods for low-resource languages, demonstrating the effectiveness of a dictionary-based approach for cross-lingual vocabulary transfer.
Can You Share Your Story? Modeling Clients’ Metacognition and Openness for LLM Therapist Evaluation
PDF ↗Understanding clients’ thoughts and beliefs is fundamental in counseling, yet current evaluations of LLM therapists often fail to assess this ability. Existing evaluation methods rely on client simulators that clearly disclose internal states to the therapist, making it difficult to determine whether an LLM therapist can uncover unexpressed perspectives. To address this limitation, we introduce MindVoyager, a novel evaluation framework featuring a controllable and realistic client simulator which dynamically adapts itself based on the ongoing counseling session, offering a more realistic and challenging evaluation environment. We further introduce evaluation metrics that assess the exploration ability of LLM therapists by measuring their thorough understanding of client’s beliefs and thoughts.
Prosody affects how people produce and understand language, yet studies of how it does so have been hindered by the lack of efficient tools for analyzing prosodic stress. We fine-tune OpenAI Whisper large-v2, a state-of-the-art speech recognition model, to recognize phrasal, lexical, and contrastive stress using a small, carefully annotated dataset. Our results show that Whisper can learn distinct, gender-specific stress patterns to achieve near-human and super-human accuracy in stress classification and transfer its learning from one type of stress to another, surpassing traditional machine learning models. Furthermore, we explore how acoustic context influences its performance and propose a novel black-box evaluation method for characterizing the decision boundaries used by Whisper for prosodic stress interpretation. These findings open new avenues for large-scale, automated prosody research. Models can be found at github.com/SSSohn/ProsodyBench.
Prompt-based text embedding models, which generate task-specific embeddings upon receiving tailored prompts, have recently demonstrated remarkable performance. However, their resulting embeddings often have thousands of dimensions, leading to high storage costs and increased computational costs of embedding-based operations. In this paper, we investigate how post-hoc dimensionality reduction applied to the embeddings affects the performance of various tasks that leverage these embeddings, specifically classification, clustering, retrieval, and semantic textual similarity (STS) tasks. Our experiments show that even a naive dimensionality reduction, which keeps only the first 25% of the dimensions of the embeddings, results in a very slight performance degradation, indicating that these embeddings are highly redundant. Notably, for classification and clustering, even when embeddings are reduced to less than 0.5% of the original dimensionality the performance degradation is very small. To quantitatively analyze this redundancy, we perform an analysis based on the intrinsic dimensionality and isotropy of the embeddings. Our analysis reveals that embeddings for classification and clustering, which are considered to have very high dimensional redundancy, exhibit lower intrinsic dimensionality and less isotropy compared with those for retrieval and STS.
As Al advances in text generation, human trust in Al generated content remains constrained by biases that go beyond concerns of accuracy. This study explores how bias shapes the perception of AI versus human generated content. Through three experiments involving text rephrasing, news article summarization, and persuasive writing, we investigated how human raters respond to labeled and unlabeled content. While the raters could not differentiate the two types of texts in the blind test, they overwhelmingly favored content labeled as “Human Generated,” over those labeled “AI Generated,” by a preference score of over 30%. We observed the same pattern even when the labels were deliberately swapped. This human bias against AI has broader societal and cognitive implications, as it undervalues AI performance. This study highlights the limitations of human judgment in interacting with AI and offers a foundation for improving human-AI collaboration, especially in creative fields.
In this paper, we propose the first multilingual study on definition modeling. We use monolingual dictionary data for four new languages (Spanish, French, Portuguese, and German) and perform an in-depth empirical study to test the performance of pre-trained multilingual language models on definition modeling of monosemic words when finetuned on this data. Furthermore, we use a zero-shot approach to test the multilingual capabilities of two popular chat-based Large Language Models (LLMs) in the task. Results show that multilingual language models can perform on-pair with English but cannot leverage potential cross-lingual synergies, with LLMs generally offering better performance overall. A comprehensive human evaluation of the LLM-generated definition highlights the zero and few-shot capabilities of these models in this new task, also showing their shortcomings. Finally, we show that performance on our task via BERTScore strongly correlates to the performance on multilingual LLM benchmarks, suggesting that our task offers a viable compute-constrained, stable and natural alternative to these.