Shangbin Feng

Yuyang Bai, Shangbin Feng, Vidhisha Balachandran et al. · cmu

Large language models (LLMs) demonstrate remarkable performance on knowledge-intensive tasks, suggesting that real-world knowledge is encoded in their model parameters. However, besides explorations on a few probing tasks in limited knowledge domains, it is not well understood how to evaluate LLMs' knowledge systematically and how well their knowledge abilities generalize, across a spectrum of knowledge domains and progressively complex task formats. To this end, we propose KGQuiz, a knowledge-intensive benchmark to comprehensively investigate the knowledge generalization abilities of LLMs. KGQuiz is a scalable framework constructed from triplet-based knowledge, which covers three knowledge domains and consists of five tasks with increasing complexity: true-or-false, multiple-choice QA, blank filling, factual editing, and open-ended knowledge generation. To gain a better understanding of LLMs' knowledge abilities and their generalization, we evaluate 10 open-source and black-box LLMs on the KGQuiz benchmark across the five knowledge-intensive tasks and knowledge domains. Extensive experiments demonstrate that LLMs achieve impressive performance in straightforward knowledge QA tasks, while settings and contexts requiring more complex reasoning or employing domain-specific facts still present significant challenges. We envision KGQuiz as a testbed to analyze such nuanced variations in performance across domains and task formats, and ultimately to understand, evaluate, and improve LLMs' knowledge abilities across a wide spectrum of knowledge domains and tasks.

Yike Wang, Shangbin Feng, Heng Wang et al. · berkeley, cmu

Large language models (LLMs) often encounter knowledge conflicts, scenarios where discrepancy arises between the internal parametric knowledge of LLMs and non-parametric information provided in the prompt context. In this work we ask what are the desiderata for LLMs when a knowledge conflict arises and whether existing LLMs fulfill them. We posit that LLMs should 1) identify knowledge conflicts, 2) pinpoint conflicting information segments, and 3) provide distinct answers or viewpoints in conflicting scenarios. To this end, we introduce an evaluation framework for simulating contextual knowledge conflicts and quantitatively evaluating to what extent LLMs achieve these goals. It includes diverse and complex situations of knowledge conflict, knowledge from diverse entities and domains, two synthetic conflict creation methods, and settings with progressively increasing difficulty to reflect realistic knowledge conflicts. Extensive experiments with the framework reveal that while LLMs perform well in identifying the existence of knowledge conflicts, they struggle to determine the specific conflicting knowledge and produce a response with distinct answers amidst conflicting information. To address these challenges, we propose new instruction-based approaches that augment LLMs to better achieve the three goals. Further analysis shows that abilities to tackle knowledge conflicts are greatly impacted by factors such as knowledge domain, while generating robust responses to knowledge conflict scenarios remains an open research question.

Bingbing Wen, Jihan Yao, Shangbin Feng et al. · allen-ai, uw

Abstention, the refusal of large language models (LLMs) to provide an answer, is increasingly recognized for its potential to mitigate hallucinations and enhance safety in LLM systems. In this survey, we introduce a framework to examine abstention from three perspectives: the query, the model, and human values. We organize the literature on abstention methods, benchmarks, and evaluation metrics using this framework, and discuss merits and limitations of prior work. We further identify and motivate areas for future research, such as whether abstention can be achieved as a meta-capability that transcends specific tasks or domains, and opportunities to optimize abstention abilities in specific contexts. In doing so, we aim to broaden the scope and impact of abstention methodologies in AI systems.

Heng Wang, Wenqian Zhang, Yuyang Bai et al.

Online movie review platforms are providing crowdsourced feedback for the film industry and the general public, while spoiler reviews greatly compromise user experience. Although preliminary research efforts were made to automatically identify spoilers, they merely focus on the review content itself, while robust spoiler detection requires putting the review into the context of facts and knowledge regarding movies, user behavior on film review platforms, and more. In light of these challenges, we first curate a large-scale network-based spoiler detection dataset LCS and a comprehensive and up-to-date movie knowledge base UKM. We then propose MVSD, a novel Multi-View Spoiler Detection framework that takes into account the external knowledge about movies and user activities on movie review platforms. Specifically, MVSD constructs three interconnecting heterogeneous information networks to model diverse data sources and their multi-view attributes, while we design and employ a novel heterogeneous graph neural network architecture for spoiler detection as node-level classification. Extensive experiments demonstrate that MVSD advances the state-of-the-art on two spoiler detection datasets, while the introduction of external knowledge and user interactions help ground robust spoiler detection. Our data and code are available at https://github.com/Arthur-Heng/Spoiler-Detection

Xinshun Feng, Herun Wan, Shangbin Feng et al.

Current Graph Neural Networks (GNNs) suffer from the over-smoothing problem, which results in indistinguishable node representations and low model performance with more GNN layers. Many methods have been put forward to tackle this problem in recent years. However, existing tackling over-smoothing methods emphasize model performance and neglect the over-smoothness of node representations. Additional, different approaches are applied one at a time, while there lacks an overall framework to jointly leverage multiple solutions to the over-smoothing challenge. To solve these problems, we propose GraTO, a framework based on neural architecture search to automatically search for GNNs architecture. GraTO adopts a novel loss function to facilitate striking a balance between model performance and representation smoothness. In addition to existing methods, our search space also includes DropAttribute, a novel scheme for alleviating the over-smoothing challenge, to fully leverage diverse solutions. We conduct extensive experiments on six real-world datasets to evaluate GraTo, which demonstrates that GraTo outperforms baselines in the over-smoothing metrics and achieves competitive performance in accuracy. GraTO is especially effective and robust with increasing numbers of GNN layers. Further experiments bear out the quality of node representations learned with GraTO and the effectiveness of model architecture. We make cide of GraTo available at Github (\url{https://github.com/fxsxjtu/GraTO}).

Yuhan Liu, Shangbin Feng, Xiaochuang Han et al. · cmu

In this work, we take a first step towards designing summarization systems that are faithful to the author's intent, not only the semantic content of the article. Focusing on a case study of preserving political perspectives in news summarization, we find that existing approaches alter the political opinions and stances of news articles in more than 50% of summaries, misrepresenting the intent and perspectives of the news authors. We thus propose P^3SUM, a diffusion model-based summarization approach controlled by political perspective classifiers. In P^3SUM, the political leaning of a generated summary is iteratively evaluated at each decoding step, and any drift from the article's original stance incurs a loss back-propagated to the embedding layers, steering the political stance of the summary at inference time. Extensive experiments on three news summarization datasets demonstrate that P^3SUM outperforms state-of-the-art summarization systems and large language models by up to 13.7% in terms of the success rate of stance preservation, with competitive performance on standard metrics of summarization quality. Our findings present a first analysis of preservation of pragmatic features in summarization, highlight the lacunae in existing summarization models -- that even state-of-the-art models often struggle to preserve author's intents -- and develop new summarization systems that are more faithful to author's perspectives.

Qingkai Zeng, Yuyang Bai, Zhaoxuan Tan et al.

Taxonomies play a crucial role in various applications by providing a structural representation of knowledge. The task of taxonomy expansion involves integrating emerging concepts into existing taxonomies by identifying appropriate parent concepts for these new query concepts. Previous approaches typically relied on self-supervised methods that generate annotation data from existing taxonomies. However, these methods are less effective when the existing taxonomy is small (fewer than 100 entities). In this work, we introduce CodeTaxo, a novel approach that leverages large language models through code language prompts to capture the taxonomic structure. Extensive experiments on five real-world benchmarks from different domains demonstrate that CodeTaxo consistently achieves superior performance across all evaluation metrics, significantly outperforming previous state-of-the-art methods. The code and data are available at https://github.com/QingkaiZeng/CodeTaxo-Pub.

Wenxuan Ding, Shangbin Feng, Yuhan Liu et al. · cmu

We propose Knowledge Crosswords, a geometric knowledge reasoning benchmark consisting of incomplete knowledge networks bounded by structured factual constraints, where LLMs are tasked with inferring the missing facts to meet all constraints. The novel setting of geometric knowledge reasoning necessitates new LM abilities beyond existing atomic/linear multi-hop QA, such as backtracking, verifying facts and constraints, reasoning with uncertainty, and more. Knowledge Crosswords contains 2,101 individual problems, covering diverse knowledge domains, and is further divided into three difficulty levels. We conduct extensive experiments to evaluate existing LLMs and approaches on Knowledge Crosswords. Results demonstrate that baseline approaches struggle with larger knowledge networks and semantically-equivalent entity distractors. In light of their limitations, we propose two new approaches, Staged Prompting and Verify-All, to augment LLMs' abilities for error-aware backtracking and constraint verification. Our Verify-All significantly outperforms prior methods and is more robust towards problems in the hard subset. Further analysis shows that geometric knowledge reasoning poses new challenges to LLMs' knowledge abilities, particularly in robustness towards varying option orders, complex structural constraints in knowledge networks, "none of the above" scenarios, and more.

Shangbin Feng, Zhaoxuan Tan, Herun Wan et al.

Twitter bot detection has become an increasingly important task to combat misinformation, facilitate social media moderation, and preserve the integrity of the online discourse. State-of-the-art bot detection methods generally leverage the graph structure of the Twitter network, and they exhibit promising performance when confronting novel Twitter bots that traditional methods fail to detect. However, very few of the existing Twitter bot detection datasets are graph-based, and even these few graph-based datasets suffer from limited dataset scale, incomplete graph structure, as well as low annotation quality. In fact, the lack of a large-scale graph-based Twitter bot detection benchmark that addresses these issues has seriously hindered the development and evaluation of novel graph-based bot detection approaches. In this paper, we propose TwiBot-22, a comprehensive graph-based Twitter bot detection benchmark that presents the largest dataset to date, provides diversified entities and relations on the Twitter network, and has considerably better annotation quality than existing datasets. In addition, we re-implement 35 representative Twitter bot detection baselines and evaluate them on 9 datasets, including TwiBot-22, to promote a fair comparison of model performance and a holistic understanding of research progress. To facilitate further research, we consolidate all implemented codes and datasets into the TwiBot-22 evaluation framework, where researchers could consistently evaluate new models and datasets. The TwiBot-22 Twitter bot detection benchmark and evaluation framework are publicly available at https://twibot22.github.io/

Zhaoxuan Tan, Zilong Chen, Shangbin Feng et al.

Knowledge Graph Embeddings (KGE) aim to map entities and relations to low dimensional spaces and have become the \textit{de-facto} standard for knowledge graph completion. Most existing KGE methods suffer from the sparsity challenge, where it is harder to predict entities that appear less frequently in knowledge graphs. In this work, we propose a novel framework KRACL to alleviate the widespread sparsity in KGs with graph context and contrastive learning. Firstly, we propose the Knowledge Relational Attention Network (KRAT) to leverage the graph context by simultaneously projecting neighboring triples to different latent spaces and jointly aggregating messages with the attention mechanism. KRAT is capable of capturing the subtle semantic information and importance of different context triples as well as leveraging multi-hop information in knowledge graphs. Secondly, we propose the knowledge contrastive loss by combining the contrastive loss with cross entropy loss, which introduces more negative samples and thus enriches the feedback to sparse entities. Our experiments demonstrate that KRACL achieves superior results across various standard knowledge graph benchmarks, especially on WN18RR and NELL-995 which have large numbers of low in-degree entities. Extensive experiments also bear out KRACL's effectiveness in handling sparse knowledge graphs and robustness against noisy triples.

Zhenyu Lei, Herun Wan, Wenqian Zhang et al.

Twitter bots are automatic programs operated by malicious actors to manipulate public opinion and spread misinformation. Research efforts have been made to automatically identify bots based on texts and networks on social media. Existing methods only leverage texts or networks alone, and while few works explored the shallow combination of the two modalities, we hypothesize that the interaction and information exchange between texts and graphs could be crucial for holistically evaluating bot activities on social media. In addition, according to a recent survey (Cresci, 2020), Twitter bots are constantly evolving while advanced bots steal genuine users' tweets and dilute their malicious content to evade detection. This results in greater inconsistency across the timeline of novel Twitter bots, which warrants more attention. In light of these challenges, we propose BIC, a Twitter Bot detection framework with text-graph Interaction and semantic Consistency. Specifically, in addition to separately modeling the two modalities on social media, BIC employs a text-graph interaction module to enable information exchange across modalities in the learning process. In addition, given the stealing behavior of novel Twitter bots, BIC proposes to model semantic consistency in tweets based on attention weights while using it to augment the decision process. Extensive experiments demonstrate that BIC consistently outperforms state-of-the-art baselines on two widely adopted datasets. Further analyses reveal that text-graph interactions and modeling semantic consistency are essential improvements and help combat bot evolution.

Wenqian Zhang, Shangbin Feng, Zilong Chen et al.

Political perspective detection has become an increasingly important task that can help combat echo chambers and political polarization. Previous approaches generally focus on leveraging textual content to identify stances, while they fail to reason with background knowledge or leverage the rich semantic and syntactic textual labels in news articles. In light of these limitations, we propose KCD, a political perspective detection approach to enable multi-hop knowledge reasoning and incorporate textual cues as paragraph-level labels. Specifically, we firstly generate random walks on external knowledge graphs and infuse them with news text representations. We then construct a heterogeneous information network to jointly model news content as well as semantic, syntactic and entity cues in news articles. Finally, we adopt relational graph neural networks for graph-level representation learning and conduct political perspective detection. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods on two benchmark datasets. We further examine the effect of knowledge walks and textual cues and how they contribute to our approach's data efficiency.

Shangbin Feng, Zhaoxuan Tan, Zilong Chen et al.

Modeling the ideological perspectives of political actors is an essential task in computational political science with applications in many downstream tasks. Existing approaches are generally limited to textual data and voting records, while they neglect the rich social context and valuable expert knowledge for holistic ideological analysis. In this paper, we propose \textbf{PAR}, a \textbf{P}olitical \textbf{A}ctor \textbf{R}epresentation learning framework that jointly leverages social context and expert knowledge. Specifically, we retrieve and extract factual statements about legislators to leverage social context information. We then construct a heterogeneous information network to incorporate social context and use relational graph neural networks to learn legislator representations. Finally, we train PAR with three objectives to align representation learning with expert knowledge, model ideological stance consistency, and simulate the echo chamber phenomenon. Extensive experiments demonstrate that PAR is better at augmenting political text understanding and successfully advances the state-of-the-art in political perspective detection and roll call vote prediction. Further analysis proves that PAR learns representations that reflect the political reality and provide new insights into political behavior.

Shangbin Feng, Zhaoxuan Tan, Wenqian Zhang et al.

With the advent of pretrained language models (LMs), increasing research efforts have been focusing on infusing commonsense and domain-specific knowledge to prepare LMs for downstream tasks. These works attempt to leverage knowledge graphs, the de facto standard of symbolic knowledge representation, along with pretrained LMs. While existing approaches have leveraged external knowledge, it remains an open question how to jointly incorporate knowledge graphs representing varying contexts, from local (e.g., sentence), to document-level, to global knowledge, to enable knowledge-rich exchange across these contexts. Such rich contextualization can be especially beneficial for long document understanding tasks since standard pretrained LMs are typically bounded by the input sequence length. In light of these challenges, we propose KALM, a Knowledge-Aware Language Model that jointly leverages knowledge in local, document-level, and global contexts for long document understanding. KALM first encodes long documents and knowledge graphs into the three knowledge-aware context representations. It then processes each context with context-specific layers, followed by a context fusion layer that facilitates knowledge exchange to derive an overarching document representation. Extensive experiments demonstrate that KALM achieves state-of-the-art performance on six long document understanding tasks and datasets. Further analyses reveal that the three knowledge-aware contexts are complementary and they all contribute to model performance, while the importance and information exchange patterns of different contexts vary with respect to different tasks and datasets.

Claudia Flores-Saviaga, Shangbin Feng, Saiph Savage

The limited information (data voids) on political topics relevant to underrepresented communities has facilitated the spread of disinformation. Independent journalists who combat disinformation in underrepresented communities have reported feeling overwhelmed because they lack the tools necessary to make sense of the information they monitor and address the data voids. In this paper, we present a system to identify and address political data voids within underrepresented communities. Armed with an interview study, indicating that the independent news media has the potential to address them, we designed an intelligent collaborative system, called Datavoidant. Datavoidant uses state-of-the-art machine learning models and introduces a novel design space to provide independent journalists with a collective understanding of data voids to facilitate generating content to cover the voids. We performed a user interface evaluation with independent news media journalists (N=22). These journalists reported that Datavoidant's features allowed them to more rapidly while easily having a sense of what was taking place in the information ecosystem to address the data voids. They also reported feeling more confident about the content they created and the unique perspectives they had proposed to cover the voids. We conclude by discussing how Datavoidant enables a new design space wherein individuals can collaborate to make sense of their information ecosystem and actively devise strategies to prevent disinformation.

Shujie Yang, Binchi Zhang, Shangbin Feng et al.

Graph anomaly detection on attributed networks has become a prevalent research topic due to its broad applications in many influential domains. In real-world scenarios, nodes and edges in attributed networks usually display distinct heterogeneity, i.e. attributes of different types of nodes show great variety, different types of relations represent diverse meanings. Anomalies usually perform differently from the majority in various perspectives of heterogeneity in these networks. However, existing graph anomaly detection approaches do not leverage heterogeneity in attributed networks, which is highly related to anomaly detection. In light of this problem, we propose AHEAD: a heterogeneity-aware unsupervised graph anomaly detection approach based on the encoder-decoder framework. Specifically, for the encoder, we design three levels of attention, i.e. attribute level, node type level, and edge level attentions to capture the heterogeneity of network structure, node properties and information of a single node, respectively. In the decoder, we exploit structure, attribute, and node type reconstruction terms to obtain an anomaly score for each node. Extensive experiments show the superiority of AHEAD on several real-world heterogeneous information networks compared with the state-of-arts in the unsupervised setting. Further experiments verify the effectiveness and robustness of our triple attention, model backbone, and decoder in general.

Jillian Fisher, Shangbin Feng, Robert Aron et al. · uw

As modern large language models (LLMs) become integral to everyday tasks, concerns about their inherent biases and their potential impact on human decision-making have emerged. While bias in models are well-documented, less is known about how these biases influence human decisions. This paper presents two interactive experiments investigating the effects of partisan bias in LLMs on political opinions and decision-making. Participants interacted freely with either a biased liberal, biased conservative, or unbiased control model while completing these tasks. We found that participants exposed to partisan biased models were significantly more likely to adopt opinions and make decisions which matched the LLM's bias. Even more surprising, this influence was seen when the model bias and personal political partisanship of the participant were opposite. However, we also discovered that prior knowledge of AI was weakly correlated with a reduction of the impact of the bias, highlighting the possible importance of AI education for robust mitigation of bias effects. Our findings not only highlight the critical effects of interacting with biased LLMs and its ability to impact public discourse and political conduct, but also highlights potential techniques for mitigating these risks in the future.

Zhuofeng Li, Haoxiang Zhang, Cong Wei et al.

Modern retrieval systems, whether lexical or semantic, expose a corpus through a fixed similarity interface that compresses access into a single top-k retrieval step before reasoning. This abstraction is efficient, but for agentic search, it becomes a bottleneck: exact lexical constraints, sparse clue conjunctions, local context checks, and multi-step hypothesis refinement are difficult to implement by calling a conventional off-the-shelf retriever, and evidence filtered out early cannot be recovered by stronger downstream reasoning. Agentic tasks further exacerbate this limitation because they require agents to orchestrate multiple steps, including discovering intermediate entities, combining weak clues, and revising the plan after observing partial evidence. To tackle the limitation, we study direct corpus interaction (DCI), where an agent searches the raw corpus directly with general-purpose terminal tools (e.g., grep, file reads, shell commands, lightweight scripts), without any embedding model, vector index, or retrieval API. This approach requires no offline indexing and adapts naturally to evolving local corpora. Across IR benchmarks and end-to-end agentic search tasks, this simple setup substantially outperforms strong sparse, dense, and reranking baselines on several BRIGHT and BEIR datasets, and attains strong accuracy on BrowseComp-Plus and multi-hop QA without relying on any conventional semantic retriever. Our results indicate that as language agents become stronger, retrieval quality depends not only on reasoning ability but also on the resolution of the interface through which the model interacts with the corpus, with which DCI opens a broader interface-design space for agentic search.

Haojin Wang, Yike Wang, Shangbin Feng et al.

Large reasoning models (LRMs) achieve strong performance by producing long chains of thought, but their inference costs are high and often generate redundant reasoning. Small language models (SLMs) are far more efficient, yet struggle on multi-step reasoning tasks. A natural idea is to let a large model guide a small one at inference time as a mentor, yet existing collaboration methods often promote imitation, resulting in verbose reasoning without consistent error correction. We propose MentorCollab, an inference-time collaboration method in which an LRM selectively and sparsely guides an SLM, rather than taking over generation. At randomly sampled token positions, we probe for divergences between the two models and use a lightweight verifier to decide whether the SLM should follow a short lookahead segment from its mentor or continue on its own. Across 15 SLM--LRM pairs and 3 domains (math reasoning, general knowledge, and commonsense reasoning), our method improves performance in 12 settings, with average gains of 3.0% and up to 8.0%, while adopting only having 18.4% tokens generated by the expensive mentor model on average. We find that short segments and selective probing are sufficient for effective collaboration. Our results show that selective inference-time guidance restores large-model reasoning ability without substantial inference overhead.

Shangbin Feng, Yuyang Bai, Ziyuan Yang et al.

Advancing beyond single monolithic language models (LMs), recent research increasingly recognizes the importance of model collaboration, where multiple LMs collaborate, compose, and complement each other. Existing research on this topic has mostly been disparate and disconnected, from different research communities, and lacks rigorous comparison. To consolidate existing research and establish model collaboration as a school of thought, we present MoCo: a one-stop Python library of executing, benchmarking, and comparing model collaboration algorithms at scale. MoCo features 26 model collaboration methods, spanning diverse levels of cross-model information exchange such as routing, text, logit, and model parameters. MoCo integrates 25 evaluation datasets spanning reasoning, QA, code, safety, and more, while users could flexibly bring their own data. Extensive experiments with MoCo demonstrate that most collaboration strategies outperform models without collaboration in 61.0% of (model, data) settings on average, with the most effective methods outperforming by up to 25.8%. We further analyze the scaling of model collaboration strategies, the training/inference efficiency of diverse methods, highlight that the collaborative system solves problems where single LMs struggle, and discuss future work in model collaboration, all made possible by MoCo. We envision MoCo as a valuable toolkit to facilitate and turbocharge the quest for an open, modular, decentralized, and collaborative AI future.

Shangbin Feng, Kishan Panaganti, Yulia Tsvetkov et al.

Model collaboration -- systems where multiple language models (LMs) collaborate -- combines the strengths of diverse models with cost in loading multiple LMs. We improve efficiency while preserving the strengths of collaboration by distilling collaborative patterns into a single model, where the model is trained on the outputs of the model collaboration system. At inference time, only the distilled model is employed: it imitates the collaboration while only incurring the cost of a single model. Furthermore, we propose the single-multi evolution loop: multiple LMs collaborate, each distills from the collaborative outputs, and these post-distillation improved LMs collaborate again, forming a collective evolution ecosystem where models evolve and self-improve by interacting with an environment of other models. Extensive experiments with 7 collaboration strategies and 15 tasks (QA, reasoning, factuality, etc.) demonstrate that: 1) individual models improve by 8.0% on average, absorbing the strengths of collaboration while reducing the cost to a single model; 2) the collaboration also benefits from the stronger and more synergistic LMs after distillation, improving over initial systems without evolution by 14.9% on average. Analysis reveals that the single-multi evolution loop outperforms various existing evolutionary AI methods, is compatible with diverse model/collaboration/distillation settings, and helps solve problems where the initial model/system struggles to.

Ziyuan Yang, Wenxuan Ding, Shangbin Feng et al.

Language models (LMs) are increasingly used in collaboration: multiple LMs trained by different parties collaborate through routing systems, multi-agent debate, model merging, and more. Critical safety risks remain in this decentralized paradigm: what if some of the models in multi-LLM systems are compromised or malicious? We first quantify the impact of malicious models by engineering four categories of malicious LMs, plug them into four types of popular model collaboration systems, and evaluate the compromised system across 10 datasets. We find that malicious models have a severe impact on the multi-LLM systems, especially for reasoning and safety domains where performance is lowered by 7.12% and 7.94% on average. We then propose mitigation strategies to alleviate the impact of malicious components, by employing external supervisors that oversee model collaboration to disable/mask them out to reduce their influence. On average, these strategies recover 95.31% of the initial performance, while making model collaboration systems fully resistant to malicious models remains an open research question.

Shangbin Feng, Weijia Shi, Yike Wang et al. · berkeley, cmu

Despite efforts to expand the knowledge of large language models (LLMs), knowledge gaps -- missing or outdated information in LLMs -- might always persist given the evolving nature of knowledge. In this work, we study approaches to identify LLM knowledge gaps and abstain from answering questions when knowledge gaps are present. We first adapt existing approaches to model calibration or adaptation through fine-tuning/prompting and analyze their ability to abstain from generating low-confidence outputs. Motivated by their failures in self-reflection and over-reliance on held-out sets, we propose two novel approaches that are based on model collaboration, i.e., LLMs probing other LLMs for knowledge gaps, either cooperatively or competitively. Extensive experiments with three LLMs on four QA tasks featuring diverse knowledge domains demonstrate that both cooperative and competitive approaches to unveiling LLM knowledge gaps achieve up to 19.3% improvements on abstain accuracy against the strongest baseline. Further analysis reveals that our proposed mechanisms could help identify failure cases in retrieval augmentation and pinpoint knowledge gaps in multi-hop reasoning.

Shangbin Feng, Taylor Sorensen, Yuhan Liu et al.

While existing alignment paradigms have been integral in developing large language models (LLMs), LLMs often learn an averaged human preference and struggle to model diverse preferences across cultures, demographics, and communities. We propose Modular Pluralism, a modular framework based on multi-LLM collaboration for pluralistic alignment: it "plugs into" a base LLM a pool of smaller but specialized community LMs, where models collaborate in distinct modes to flexibility support three modes of pluralism: Overton, steerable, and distributional. Modular Pluralism is uniquely compatible with black-box LLMs and offers the modular control of adding new community LMs for previously underrepresented communities. We evaluate Modular Pluralism with six tasks and four datasets featuring questions/instructions with value-laden and perspective-informed responses. Extensive experiments demonstrate that Modular Pluralism advances the three pluralism objectives across six black-box and open-source LLMs. Further analysis reveals that LLMs are generally faithful to the inputs from smaller community LLMs, allowing seamless patching by adding a new community LM to better cover previously underrepresented communities.

Heng Wang, Shangbin Feng, Tianxing He et al.

Large language models (LLMs) are increasingly adopted for a variety of tasks with implicit graphical structures, such as planning in robotics, multi-hop question answering or knowledge probing, structured commonsense reasoning, and more. While LLMs have advanced the state-of-the-art on these tasks with structure implications, whether LLMs could explicitly process textual descriptions of graphs and structures, map them to grounded conceptual spaces, and perform structured operations remains underexplored. To this end, we propose NLGraph (Natural Language Graph), a comprehensive benchmark of graph-based problem solving designed in natural language. NLGraph contains 29,370 problems, covering eight graph reasoning tasks with varying complexity from simple tasks such as connectivity and shortest path up to complex problems such as maximum flow and simulating graph neural networks. We evaluate LLMs (GPT-3/4) with various prompting approaches on the NLGraph benchmark and find that 1) language models do demonstrate preliminary graph reasoning abilities, 2) the benefit of advanced prompting and in-context learning diminishes on more complex graph problems, while 3) LLMs are also (un)surprisingly brittle in the face of spurious correlations in graph and problem settings. We then propose Build-a-Graph Prompting and Algorithmic Prompting, two instruction-based approaches to enhance LLMs in solving natural language graph problems. Build-a-Graph and Algorithmic prompting improve the performance of LLMs on NLGraph by 3.07% to 16.85% across multiple tasks and settings, while how to solve the most complicated graph reasoning tasks in our setup with language models remains an open research question. The NLGraph benchmark and evaluation code are available at https://github.com/Arthur-Heng/NLGraph.

Herun Wan, Shangbin Feng, Zhaoxuan Tan et al.

Large language models are limited by challenges in factuality and hallucinations to be directly employed off-the-shelf for judging the veracity of news articles, where factual accuracy is paramount. In this work, we propose DELL that identifies three key stages in misinformation detection where LLMs could be incorporated as part of the pipeline: 1) LLMs could \emph{generate news reactions} to represent diverse perspectives and simulate user-news interaction networks; 2) LLMs could \emph{generate explanations} for proxy tasks (e.g., sentiment, stance) to enrich the contexts of news articles and produce experts specializing in various aspects of news understanding; 3) LLMs could \emph{merge task-specific experts} and provide an overall prediction by incorporating the predictions and confidence scores of varying experts. Extensive experiments on seven datasets with three LLMs demonstrate that DELL outperforms state-of-the-art baselines by up to 16.8\% in macro f1-score. Further analysis reveals that the generated reactions and explanations are greatly helpful in misinformation detection, while our proposed LLM-guided expert merging helps produce better-calibrated predictions.

Shangbin Feng, Herun Wan, Ningnan Wang et al.

Social media bot detection has always been an arms race between advancements in machine learning bot detectors and adversarial bot strategies to evade detection. In this work, we bring the arms race to the next level by investigating the opportunities and risks of state-of-the-art large language models (LLMs) in social bot detection. To investigate the opportunities, we design novel LLM-based bot detectors by proposing a mixture-of-heterogeneous-experts framework to divide and conquer diverse user information modalities. To illuminate the risks, we explore the possibility of LLM-guided manipulation of user textual and structured information to evade detection. Extensive experiments with three LLMs on two datasets demonstrate that instruction tuning on merely 1,000 annotated examples produces specialized LLMs that outperform state-of-the-art baselines by up to 9.1% on both datasets, while LLM-guided manipulation strategies could significantly bring down the performance of existing bot detectors by up to 29.6% and harm the calibration and reliability of bot detection systems.

Yichen Wang, Shangbin Feng, Abe Bohan Hou et al. · berkeley

The widespread use of large language models (LLMs) is increasing the demand for methods that detect machine-generated text to prevent misuse. The goal of our study is to stress test the detectors' robustness to malicious attacks under realistic scenarios. We comprehensively study the robustness of popular machine-generated text detectors under attacks from diverse categories: editing, paraphrasing, prompting, and co-generating. Our attacks assume limited access to the generator LLMs, and we compare the performance of detectors on different attacks under different budget levels. Our experiments reveal that almost none of the existing detectors remain robust under all the attacks, and all detectors exhibit different loopholes. Averaging all detectors, the performance drops by 35% across all attacks. Further, we investigate the reasons behind these defects and propose initial out-of-the-box patches to improve robustness.

Qingkai Zeng, Yuyang Bai, Zhaoxuan Tan et al.

Automatic taxonomy induction is crucial for web search, recommendation systems, and question answering. Manual curation of taxonomies is expensive in terms of human effort, making automatic taxonomy construction highly desirable. In this work, we introduce Chain-of-Layer which is an in-context learning framework designed to induct taxonomies from a given set of entities. Chain-of-Layer breaks down the task into selecting relevant candidate entities in each layer and gradually building the taxonomy from top to bottom. To minimize errors, we introduce the Ensemble-based Ranking Filter to reduce the hallucinated content generated at each iteration. Through extensive experiments, we demonstrate that Chain-of-Layer achieves state-of-the-art performance on four real-world benchmarks.

Shangbin Feng, Wenxuan Ding, Alisa Liu et al. · berkeley, mit

This position paper argues that in many realistic (i.e., complex, contextualized, subjective) scenarios, one LLM is not enough to produce a reliable output. We challenge the status quo of relying solely on a single general-purpose LLM and argue for multi-LLM collaboration to better represent the extensive diversity of data, skills, and people. We first posit that a single LLM underrepresents real-world data distributions, heterogeneous skills, and pluralistic populations, and that such representation gaps cannot be trivially patched by further training a single LLM. We then organize existing multi-LLM collaboration methods into a hierarchy, based on the level of access and information exchange, ranging from API-level, text-level, logit-level, to weight-level collaboration. Based on these methods, we highlight how multi-LLM collaboration addresses challenges that a single LLM struggles with, such as reliability, democratization, and pluralism. Finally, we identify the limitations of existing multi-LLM methods and motivate future work. We envision multi-LLM collaboration as an essential path toward compositional intelligence and collaborative AI development.

Shangbin Feng, Zifeng Wang, Yike Wang et al. · berkeley

We propose Model Swarms, a collaborative search algorithm to adapt LLMs via swarm intelligence, the collective behavior guiding individual systems. Specifically, Model Swarms starts with a pool of LLM experts and a utility function. Guided by the best-found checkpoints across models, diverse LLM experts collaboratively move in the weight space and optimize a utility function representing model adaptation objectives. Compared to existing model composition approaches, Model Swarms offers tuning-free model adaptation, works in low-data regimes with as few as 200 examples, and does not require assumptions about specific experts in the swarm or how they should be composed. Extensive experiments demonstrate that Model Swarms could flexibly adapt LLM experts to a single task, multi-task domains, reward models, as well as diverse human interests, improving over 12 model composition baselines by up to 21.0% across tasks and contexts. Further analysis reveals that LLM experts discover previously unseen capabilities in initial checkpoints and that Model Swarms enable the weak-to-strong transition of experts through the collaborative search process.

Shangbin Feng, Zifeng Wang, Palash Goyal et al. · berkeley

We propose Heterogeneous Swarms, an algorithm to design multi-LLM systems by jointly optimizing model roles and weights. We represent multi-LLM systems as directed acyclic graphs (DAGs) of LLMs with topological message passing for collaborative generation. Given a pool of LLM experts and a utility function, Heterogeneous Swarms employs two iterative steps: role-step and weight-step. For role-step, we interpret model roles as learning a DAG that specifies the flow of inputs and outputs between LLMs. Starting from a swarm of random continuous adjacency matrices, we decode them into discrete DAGs, call the LLMs in topological order, evaluate on the utility function (e.g. accuracy on a task), and optimize the adjacency matrices with particle swarm optimization based on the utility score. For weight-step, we assess the contribution of individual LLMs in the multi-LLM systems and optimize model weights with swarm intelligence. We propose JFK-score to quantify the individual contribution of each LLM in the best-found DAG of the role-step, then optimize model weights with particle swarm optimization based on the JFK-score. Experiments demonstrate that Heterogeneous Swarms outperforms 15 role- and/or weight-based baselines by 18.5% on average across 12 tasks. Further analysis reveals that Heterogeneous Swarms discovers multi-LLM systems with heterogeneous model roles and substantial collaborative gains, and benefits from the diversity of language models.

Jillian Fisher, Ruth E. Appel, Chan Young Park et al. · uw

AI systems often exhibit political bias, influencing users' opinions and decisions. While political neutrality-defined as the absence of bias-is often seen as an ideal solution for fairness and safety, this position paper argues that true political neutrality is neither feasible nor universally desirable due to its subjective nature and the biases inherent in AI training data, algorithms, and user interactions. However, inspired by Joseph Raz's philosophical insight that "neutrality [...] can be a matter of degree" (Raz, 1986), we argue that striving for some neutrality remains essential for promoting balanced AI interactions and mitigating user manipulation. Therefore, we use the term "approximation" of political neutrality to shift the focus from unattainable absolutes to achievable, practical proxies. We propose eight techniques for approximating neutrality across three levels of conceptualizing AI, examining their trade-offs and implementation strategies. In addition, we explore two concrete applications of these approximations to illustrate their practicality. Finally, we assess our framework on current large language models (LLMs) at the output level, providing a demonstration of how it can be evaluated. This work seeks to advance nuanced discussions of political neutrality in AI and promote the development of responsible, aligned language models.

Yuwei Zhang, Wenhao Yu, Shangbin Feng et al.

Despite significant advances in large language models (LLMs), their knowledge memorization capabilities remain underexplored, due to the lack of standardized and high-quality test ground. In this paper, we introduce a novel, real-world and large-scale knowledge injection benchmark that evolves continuously over time without requiring human intervention. Specifically, we propose WikiDYK, which leverages recently-added and human-written facts from Wikipedia's "Did You Know..." entries. These entries are carefully selected by expert Wikipedia editors based on criteria such as verifiability and clarity. Each entry is converted into multiple question-answer pairs spanning diverse task formats from easy cloze prompts to complex multi-hop questions. WikiDYK contains 12,290 facts and 77,180 questions, which is also seamlessly extensible with future updates from Wikipedia editors. Extensive experiments using continued pre-training reveal a surprising insight: despite their prevalence in modern LLMs, Causal Language Models (CLMs) demonstrate significantly weaker knowledge memorization capabilities compared to Bidirectional Language Models (BiLMs), exhibiting a 23% lower accuracy in terms of reliability. To compensate for the smaller scales of current BiLMs, we introduce a modular collaborative framework utilizing ensembles of BiLMs as external knowledge repositories to integrate with LLMs. Experiment shows that our framework further improves the reliability accuracy by up to 29.1%.

Jihan Yao, Wenxuan Ding, Shangbin Feng et al. · allen-ai, uw

In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers? We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options? And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful? We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences. Extensive experiments with seven LLMs and eight datasets demonstrate that (1) LLMs do have preliminary capability in distinguishing various shades of wrong, achieving up to 20.9% higher performance than random guess; (2) Alignment with wrong-over-wrong preferences helps LLMs to produce less wrong and sometimes even outright correct answers, while overall improving model calibration.

Rock Yuren Pang, K. J. Kevin Feng, Shangbin Feng et al. · uw

The output quality of large language models (LLMs) can be improved via "reasoning": generating segments of chain-of-thought (CoT) content to further condition the model prior to producing user-facing output. While these chains contain valuable information, they are verbose and lack explicit organization, making them tedious to review. Moreover, they lack opportunities for user feedback, such as to remove unwanted considerations, add desired ones, or clarify unclear assumptions. We introduce Interactive Reasoning, an interaction design that visualizes chain-of-thought outputs as a hierarchy of topics and enables user review and modification. We implement interactive reasoning in Hippo, a prototype for AI-assisted decision making in the face of uncertain trade-offs. In a user study with 16 participants, we find that interactive reasoning in Hippo allows users to quickly identify and interrupt erroneous generations, efficiently steer the model towards customized responses, and better understand both model reasoning and model outputs. Our work contributes to a new paradigm that incorporates user oversight into LLM reasoning processes.

Yuru Jiang, Wenxuan Ding, Shangbin Feng et al.

We propose SPARTA ALIGNMENT, an algorithm to collectively align multiple LLMs through competition and combat. To complement a single model's lack of diversity in generation and biases in evaluation, multiple LLMs form a "sparta tribe" to compete against each other in fulfilling instructions while serving as judges for the competition of others. For each iteration, one instruction and two models are selected for a duel, the other models evaluate the two responses, and their evaluation scores are aggregated through a adapted elo-ranking based reputation system, where winners/losers of combat gain/lose weight in evaluating others. The peer-evaluated combat results then become preference pairs where the winning response is preferred over the losing one, and all models learn from these preferences at the end of each iteration. SPARTA ALIGNMENT enables the self-evolution of multiple LLMs in an iterative and collective competition process. Extensive experiments demonstrate that SPARTA ALIGNMENT outperforms initial models and 4 self-alignment baselines across 10 out of 12 tasks and datasets with 7.0% average improvement. Further analysis reveals that SPARTA ALIGNMENT generalizes more effectively to unseen tasks and leverages the expertise diversity of participating models to produce more logical, direct and informative outputs.

Yike Wang, Shangbin Feng, Yulia Tsvetkov et al. · berkeley

Large Language Models (LLMs) are increasingly used to support scientific research, but their knowledge of scientific advancements can quickly become outdated. We introduce ScienceMeter, a new framework for evaluating scientific knowledge update methods over scientific knowledge spanning the past, present, and future. ScienceMeter defines three metrics: knowledge preservation, the extent to which models' understanding of previously learned papers are preserved; knowledge acquisition, how well scientific claims from newly introduced papers are acquired; and knowledge projection, the ability of the updated model to anticipate or generalize to related scientific claims that may emerge in the future. Using ScienceMeter, we examine the scientific knowledge of LLMs on claim judgment and generation tasks across a curated dataset of 15,444 scientific papers and 30,888 scientific claims from ten domains including medicine, biology, materials science, and computer science. We evaluate five representative knowledge update approaches including training- and inference-time methods. With extensive experiments, we find that the best-performing knowledge update methods can preserve only 85.9% of existing knowledge, acquire 71.7% of new knowledge, and project 37.7% of future knowledge. Inference-based methods work for larger models, whereas smaller models require training to achieve comparable performance. Cross-domain analysis reveals that performance on these objectives is correlated. Even when applying on specialized scientific LLMs, existing knowledge update methods fail to achieve these objectives collectively, underscoring that developing robust scientific knowledge update mechanisms is both crucial and challenging.

Varich Boonsanong, Vidhisha Balachandran, Xiaochuang Han et al. · allen-ai, cmu

With the widespread consumption of AI-generated content, there has been an increased focus on developing automated tools to verify the factual accuracy of such content. However, prior research and tools developed for fact verification treat it as a binary classification or a linear regression problem. Although this is a useful mechanism as part of automatic guardrails in systems, we argue that such tools lack transparency in the prediction reasoning and diversity in source evidence to provide a trustworthy user experience. We develop Facts&Evidence - an interactive and transparent tool for user-driven verification of complex text. The tool facilitates the intricate decision-making involved in fact-verification, presenting its users a breakdown of complex input texts to visualize the credibility of individual claims along with an explanation of model decisions and attribution to multiple, diverse evidence sources. Facts&Evidence aims to empower consumers of machine-generated text and give them agency to understand, verify, selectively trust and use such text.

Yiwei Yang, Chung Peng Lee, Shangbin Feng et al.

Finetuning can cause spurious correlations to arise between non-essential features and the target labels, but benchmarks to study these effects involve contrived settings and narrow tasks. In contrast, we consider spurious correlations in multi-modal Large Vision Language Models (LVLMs) pretrained on extensive and diverse datasets without explicit task supervision. We develop a benchmark by sourcing GPT-4o errors on real-world visual-question-answering (VQA) benchmarks, then curating a subset through LVLM-human annotation and synthetic counterfactual evaluation to identify errors caused by spurious correlations. This process yields SpuriVerse, a novel benchmark comprised of 124 distinct types of spurious correlations extracted from real-world datasets, each containing 1 realistic and 10 synthetic VQA samples for a total of 1364 multiple choice questions. We evaluate 15 open and closed-source LVLMs on SpuriVerse, finding that even state-of-the-art closed-source models struggle significantly, achieving at best only 37.1% accuracy. Fine-tuning on synthetic examples that emphasize the spurious correlation improves performance to 78.40%, suggesting that training on diverse spurious patterns generalizes to unseen situations: models appear to learn to avoid "shortcuts" and attend to the overall image context.

Zifeng Zhu, Shangbin Feng, Herun Wan et al.

We propose GuessBench, a novel benchmark that evaluates Vision Language Models (VLMs) on modeling the pervasive, noisy, and pluralistic human creativity. GuessBench sources data from "Guess the Build", an online multiplayer Minecraft minigame where one player constructs a Minecraft build given a concept (e.g. caterpillar) and others try to guess it with natural language hints, presenting a pristine testbed for sensemaking creativity in the wild with VLMs acting as guessers. We curate 1500 images from the actual gameplay and design 2000 problems spanning static and dynamic image settings, natural language hints of varying completeness, and more. Extensive experiments with six open/API VLMs and five reasoning enhancement approaches demonstrate that GuessBench presents a uniquely challenging task in creativity modeling: even the start-of-the-art GPT-4o is incorrect on 34% of instances, while we observe a huge performance gap (13.87% vs. 53.93% on average) between open and API models. When used as a resource to improve VLMs, fine-tuning on the reasoning traces for GuessBench problems improves visual perception tasks by 15.36% on average. Further analysis reveals that VLM performance in creativity sensemaking correlates with the frequency of the concept in training data, while the accuracy drops sharply for concepts in underrepresented cultural contexts and low-resource languages.

Jihan Yao, Yushi Hu, Yujie Yi et al. · allen-ai, uw

Automatically evaluating multimodal generation presents a significant challenge, as automated metrics often struggle to align reliably with human evaluation, especially for complex tasks that involve multiple modalities. To address this, we present MMMG, a comprehensive and human-aligned benchmark for multimodal generation across 4 modality combinations (image, audio, interleaved text and image, interleaved text and audio), with a focus on tasks that present significant challenges for generation models, while still enabling reliable automatic evaluation through a combination of models and programs. MMMG encompasses 49 tasks (including 29 newly developed ones), each with a carefully designed evaluation pipeline, and 937 instructions to systematically assess reasoning, controllability, and other key capabilities of multimodal generation models. Extensive validation demonstrates that MMMG is highly aligned with human evaluation, achieving an average agreement of 94.3%. Benchmarking results on 24 multimodal generation models reveal that even though the state-of-the-art model, GPT Image, achieves 78.3% accuracy for image generation, it falls short on multimodal reasoning and interleaved generation. Furthermore, results suggest considerable headroom for improvement in audio generation, highlighting an important direction for future research.

Shangbin Feng, Wenhao Yu, Yike Wang et al. · berkeley

Alignment training has tradeoffs: it helps language models (LMs) gain in reasoning and instruction following but might lose out on skills such as creativity and calibration, where unaligned base models are better at. We aim to make the best of both worlds through model collaboration, where different models in the training pipeline collaborate and complement each other. Since LM responses feature interleaving skills that favor different models, we propose Switch Generation, where pretrained and aligned model versions take turns to ``speak'' in a response sequence. Specifically, we train a switcher LM by learning from outcomes of choosing different models to generate the next segment across diverse queries and contexts. At inference time, the switcher LM guides different model checkpoints to dynamically generate the next segment where their strengths are most needed. Extensive experiments with 8 model collaboration baselines and 18 datasets show that 1) model collaboration consistently outperforms individual models on 16 out of 18 tasks, and 2) Switch Generation further outperforms baselines by 12.9% on average. Further analysis reveals that Switch Generation discovers compositional skills to solve problems where individual models struggle and generalizes to unseen models and tasks, reusing and repurposing by-products in expensive model training pipelines that are otherwise discarded.

Yizhuo Zhang, Heng Wang, Shangbin Feng et al. · tsinghua, uw

Previous research has sought to enhance the graph reasoning capabilities of LLMs by supervised fine-tuning on synthetic graph data. While these led to specialized LLMs better at solving graph algorithm problems, we don't need LLMs for shortest path: we need generalization from synthetic graph data to real-world tasks with implicit graph structures. In this work, we propose to unlock generalizable learning of graph with post-training alignment with synthetic data. We first design solution-based and process-based rewards for synthetic graph problems: instead of rigid memorizing response patterns in direct fine-tuning, we posit that post-training alignment would help LLMs grasp the essentials underlying graph reasoning and alleviate overfitting on synthetic data. We employ post-training alignment algorithms such as GRPO and DPO, aligning both off-the-shelf LLMs and LLMs fine-tuned on synthetic graph data. We then compare them against existing settings on both in-domain synthetic tasks and out-of-domain real-world tasks with implicit graph structures such as multi-hop QA, structured planning, and more. Extensive experiments demonstrate that our post-training alignment recipe leads to statistically significant improvement on 5 datasets, with an average gain of 12.9% over baseline settings. Further analysis reveals that process-based rewards consistently outperform solution-based rewards on synthetic data but not on real-world tasks, and compositionality and explainable intermediate steps remains a critical challenge even after post-training alignment.

Shangbin Feng, Yike Wang, Weijia Shi et al. · berkeley

We propose Data Swarms, an algorithm to optimize the generation of synthetic evaluation data and advance quantitative desiderata of LLM evaluation. We first train a swarm of initial data generators using existing data, and define various evaluation objectives to reflect the desired properties of evaluation (e.g., generate more difficult problems for the evaluated models) and quantitatively evaluate data generators. We then employ particle swarm optimization to optimize the swarm of data generators, where they collaboratively search through the model parameter space to find new generators that advance these objectives. We further extend it to Adversarial Swarms, where the data generator swarm generates harder data while the test taker model swarm learns from such data, co-evolving dynamically for better data and models simultaneously. Extensive experiments demonstrate that Data Swarms outperforms eight data generation baselines across five evaluation objectives, while Adversarial Swarms produce more robust learning of synthetic data and stronger generalization. Further analysis reveals that Data Swarms successfully optimizes compositions of multiple evaluation objectives and generalizes to new off-the-shelf LLMs, unseen at optimization time.

Yizhuo Zhang, Heng Wang, Shangbin Feng et al.

Large language models (LLMs) demonstrate great potential for problems with implicit graphical structures, while recent works seek to enhance the graph reasoning capabilities of LLMs through specialized instruction tuning. The resulting 'graph LLMs' are evaluated with in-distribution settings only, thus it remains underexplored whether LLMs are learning generalizable graph reasoning skills or merely memorizing patterns in the synthetic training data. To this end, we propose the NLGift benchmark, an evaluation suite of LLM graph reasoning generalization: whether LLMs could go beyond semantic, numeric, structural, reasoning patterns in the synthetic training data and improve utility on real-world graph-based tasks. Extensive experiments with two LLMs across four graph reasoning tasks demonstrate that while generalization on simple patterns (semantic, numeric) is somewhat satisfactory, LLMs struggle to generalize across reasoning and real-world patterns, casting doubt on the benefit of synthetic graph tuning for real-world tasks with underlying network structures. We explore three strategies to improve LLM graph reasoning generalization, and we find that while post-training alignment is most promising for real-world tasks, empowering LLM graph reasoning to go beyond pattern memorization remains an open research question.

Shangbin Feng, Weijia Shi, Yike Wang et al.

Multilingual LLMs often have knowledge disparities across languages, with larger gaps in under-resourced languages. Teaching LLMs to abstain in the face of knowledge gaps is thus a promising strategy to mitigate hallucinations in multilingual settings. However, previous studies on LLM abstention primarily focus on English; we find that directly applying existing solutions beyond English results in up to 20.5% performance gaps between high and low-resource languages, potentially due to LLMs' drop in calibration and reasoning beyond a few resource-rich languages. To this end, we propose strategies to enhance LLM abstention by learning from multilingual feedback, where LLMs self-reflect on proposed answers in one language by generating multiple feedback items in related languages: we show that this helps identifying the knowledge gaps across diverse languages, cultures, and communities. Extensive experiments demonstrate that our multilingual feedback approach outperforms various strong baselines, achieving up to 9.2% improvement for low-resource languages across three black-box and open models on three datasets, featuring open-book, closed-book, and commonsense QA. Further analysis reveals that multilingual feedback is both an effective and a more equitable abstain strategy to serve diverse language speakers, and cultural factors have great impact on language selection and LLM abstention behavior, highlighting future directions for multilingual and multi-cultural reliable language modeling.

Shuyue Stella Li, Vidhisha Balachandran, Shangbin Feng et al.

Users typically engage with LLMs interactively, yet most existing benchmarks evaluate them in a static, single-turn format, posing reliability concerns in interactive scenarios. We identify a key obstacle towards reliability: LLMs are trained to answer any question, even with incomplete context or insufficient knowledge. In this paper, we propose to change the static paradigm to an interactive one, develop systems that proactively ask questions to gather more information and respond reliably, and introduce an benchmark - MediQ - to evaluate question-asking ability in LLMs. MediQ simulates clinical interactions consisting of a Patient System and an adaptive Expert System; with potentially incomplete initial information, the Expert refrains from making diagnostic decisions when unconfident, and instead elicits missing details via follow-up questions. We provide a pipeline to convert single-turn medical benchmarks into an interactive format. Our results show that directly prompting state-of-the-art LLMs to ask questions degrades performance, indicating that adapting LLMs to proactive information-seeking settings is nontrivial. We experiment with abstention strategies to better estimate model confidence and decide when to ask questions, improving diagnostic accuracy by 22.3%; however, performance still lags compared to an (unrealistic in practice) upper bound with complete information upfront. Further analyses show improved interactive performance with filtering irrelevant contexts and reformatting conversations. Overall, we introduce a novel problem towards LLM reliability, an interactive MediQ benchmark and a novel question-asking system, and highlight directions to extend LLMs' information-seeking abilities in critical domains.

Shangbin Feng, Weijia Shi, Yuyang Bai et al.

By design, large language models (LLMs) are static general-purpose models, expensive to retrain or update frequently. As they are increasingly adopted for knowledge-intensive tasks, it becomes evident that these design choices lead to failures to generate factual, relevant, and up-to-date knowledge. To this end, we propose Knowledge Card, a modular framework to plug in new factual and relevant knowledge into general-purpose LLMs. We first introduce knowledge cards -- specialized language models trained on corpora from specific domains and sources. Knowledge cards serve as parametric repositories that are selected at inference time to generate background knowledge for the base LLM. We then propose three content selectors to dynamically select and retain information in documents generated by knowledge cards, specifically controlling for relevance, brevity, and factuality of outputs. Finally, we propose two complementary integration approaches to augment the base LLM with the (relevant, factual) knowledge curated from the specialized LMs. Through extensive experiments, we demonstrate that Knowledge Card achieves state-of-the-art performance on six benchmark datasets. Ultimately, Knowledge Card framework enables dynamic synthesis and updates of knowledge from diverse domains. Its modularity will ensure that relevant knowledge can be continuously updated through the collective efforts of the research community.

Shangbin Feng, Chan Young Park, Yuhan Liu et al.

Language models (LMs) are pretrained on diverse data sources, including news, discussion forums, books, and online encyclopedias. A significant portion of this data includes opinions and perspectives which, on one hand, celebrate democracy and diversity of ideas, and on the other hand are inherently socially biased. Our work develops new methods to (1) measure political biases in LMs trained on such corpora, along social and economic axes, and (2) measure the fairness of downstream NLP models trained on top of politically biased LMs. We focus on hate speech and misinformation detection, aiming to empirically quantify the effects of political (social, economic) biases in pretraining data on the fairness of high-stakes social-oriented tasks. Our findings reveal that pretrained LMs do have political leanings that reinforce the polarization present in pretraining corpora, propagating social biases into hate speech predictions and misinformation detectors. We discuss the implications of our findings for NLP research and propose future directions to mitigate unfairness.