The advanced role-playing capabilities of Large Language Models (LLMs) have enabled rich interactive scenarios, yet existing research in social interactions neglects hallucination while struggling with poor generalizability and implicit character fidelity judgments. To bridge this gap, motivated by human behaviour, we introduce a generalizable and explicit paradigm for uncovering interactive patterns of LLMs across diverse worldviews. Specifically, we first define interactive hallucination through stance transfer, then construct SHARP, a benchmark built by extracting relations from commonsense knowledge graphs and utilizing LLMs’ inherent hallucination properties to simulate multi-role interactions. Extensive experiments confirm our paradigm’s effectiveness and stability, examine the factors that influence these metrics, and challenge conventional hallucination mitigation solutions. More broadly, our work reveals a fundamental limitation in popular post-training methods for role-playing LLMs: the tendency to obscure knowledge beneath style, resulting in monotonous yet human-like behaviors—interactive hallucination.
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
Analysing the differences in how events are represented across texts, or verifying whether the language model generations hallucinate, requires the ability to systematically compare their content. To support such comparison, structured representation that captures fine-grained information plays a vital role.In particular, identifying distinct atomic facts and the discourse relations connecting them enables deeper semantic comparison. Our proposed approach combines structured discourse information extraction with a classifier, FDSpotter, for factual consistency verification. We show that adversarial discourse relations pose challenges for language models, but fine-tuning on our annotated data, DiscInfer, achieves competitive performance. Our proposed approach advances factual consistency verification by grounding in linguistic structure and decomposing it into interpretable components. We demonstrate the effectiveness of our method on the evaluation of two tasks: data-to-text generation and text summarisation. Our code and dataset will be publicly available on GitHub.
Profiling News Media for Factuality and Bias Using LLMs and the Fact-Checking Methodology of Human Experts
PDF ↗In an age characterized by the proliferation of mis- and disinformation online, it is critical to empower readers to understand the content they are reading. Important efforts in this direction rely on manual or automatic fact-checking, which can be challenging for emerging claims with limited information. Such scenarios can be handled by assessing the reliability and the political bias of the source of the claim, i.e., characterizing entire news outlets rather than individual claims or articles. This is an important but understudied research direction. While prior work has looked into linguistic and social contexts, we do not analyze individual articles or information in social media. Instead, we propose a novel methodology that emulates the criteria that professional fact-checkers use to assess the factuality and political bias of an entire outlet. Specifically, we design a variety of prompts based on these criteria and elicit responses from large language models (LLMs), which we aggregate to make predictions. In addition to demonstrating sizable improvements over strong baselines via extensive experiments with multiple LLMs, we provide an in-depth error analysis of the effect of media popularity and region on model performance. Further, we conduct an ablation study to highlight the key components of our dataset that contribute to these improvements. To facilitate future research, we released our dataset and code.
Large Language Models (LLMs) have shown strong performance in solving mathematical problems, with code-based solutions proving particularly effective. However, the best practice to leverage coding instruction data to enhance mathematical reasoning remains underexplored. This study investigates three key questions: (1) How do different coding styles of mathematical code-based rationales impact LLMs’ learning performance? (2) Can general-domain coding instructions improve performance? (3) How does integrating textual rationales with code-based ones during training enhance mathematical reasoning abilities? Our findings reveal that code-based rationales with concise comments, descriptive naming, and hardcoded solutions are beneficial, while improvements from general-domain coding instructions and textual rationales are relatively minor. Based on these insights, we propose CoinMath, a learning strategy designed to enhance mathematical reasoning by diversifying the coding styles of code-based rationales. CoinMath generates a variety of code-based rationales incorporating concise comments, descriptive naming conventions, and hardcoded solutions. Experimental results demonstrate that CoinMath significantly outperforms its baseline model, MAmmoTH, one of the SOTA math LLMs.
Evaluating Instructively Generated Statement by Large Language Models for Directional Event Causality Identification
PDF ↗This paper aims to identify directional causal relations between events, including the existence and direction of causality. Previous studies mainly adopt prompt learning paradigm to predict a causal answer word based on a Pre-trained Language Model (PLM) for causality existence identification. However, the indecision in selecting answer words from some synonyms and the confusion of indicating opposite causal directions with the same answer word raise more challenges in directional causality identification. Inspired by the strong capabilities of pre-trained Generative Language Models (GLMs) in generating responses or statements, we propose to instruct a GLM to generate causality statements and identify directional event causality by evaluating the generated statements. Specifically, we propose an Instructive Generation and Statement Evaluation method to identify both the existence and direction of causality. We first fine-tune a GLM to instructively generate causality statements based on event description inputs. Then, we evaluate the rationality of the generated statements to determine the existence and direction of event causalities. Experiments on the ESC and MAVEN datasets show that our method significantly outperforms state-of-the-art algorithms, even with fewer training data.
Language models can solve complex reasoning tasks better by learning to generate rationales for their predictions. Often these models know how to solve a task but their auto-regressive decoding nature leads to incorrect results if started incorrectly. We observe that smaller models in particular, when corrected, can solve a task that they would otherwise struggle with. We demonstrate this phenomenon by using a larger model to guide smaller models, which leads to significantly improved performance (up to +24 points on the GSM8K dataset by 7B models). To assist smaller models in initiating the starting step, we propose QuestCoT, where a smaller model first asks how to start before proceeding with a chain of reasoning. On various multistep mathematical reasoning datasets over multiple smaller models, we show that getting the start right can lead to significant performance gains across all models (gains of up to +6 points on GSM8K, +9 on SVAMP, +5 on ASDiv, and +7 on MultiArith).
When different groups’ values differ, one approach to model alignment is to steer models at inference time towards each group’s preferences. However, techniques like in-context learning only consider similarity when drawing few-shot examples and not cross-group differences in values. We propose SPICA, a framework that accounts for group-level differences during in-context example retrieval. SPICA introduces three designs: scenario banks, group-informed retrieval metrics, and in-context alignment prompts. From an evaluation of SPICA on an alignment task collecting inputs from four demographic groups (n = 544), our metrics retrieve in-context examples that more closely match observed preferences, with the best prompt configuration using multiple contrastive responses to demonstrate examples. In an end-to-end evaluation (n = 120), we observe that SPICA is higher rated than similarity-based retrieval, with groups seeing up to a +0.16 point improvement on a 5 point scale. Additionally, gains from SPICA were more uniform, with all groups benefiting from alignment rather than only some. Finally, we find that while a group-agnostic approach can align to aggregated values, it is not most suited for divergent groups.
Accurately assessing internal human states is key to understanding preferences, offering personalized services, and identifying challenges in real-world applications. Originating from psychometrics, adaptive testing has become the mainstream method for human measurement and has now been widely applied in education, healthcare, sports, and sociology. It customizes assessments by selecting the fewest test questions . However, current adaptive testing methods face several challenges. The mechanized nature of most algorithms leads to guessing behavior and difficulties with open-ended questions. Additionally, subjective assessments suffer from noisy response data and coarse-grained test outputs, further limiting their effectiveness. To move closer to an ideal adaptive testing process, we propose TestAgent, a large language model (LLM)-powered agent designed to enhance adaptive testing through interactive engagement. This is the first application of LLMs in adaptive testing. TestAgent supports personalized question selection, captures test-takers’ responses and anomalies, and provides precise outcomes through dynamic, conversational interactions. Experiments on psychological, educational, and lifestyle assessments show our approach achieves more accurate results with 20% fewer questions than state-of-the-art baselines, and testers preferred it in speed, smoothness, and other dimensions.
Comparing Bad Apples to Good Oranges Aligning Large Language Models via Joint Preference Optimization
PDF ↗A common technique for aligning large language models (LLMs) relies on acquiring human preferences by comparing multiple generations conditioned on a fixed context. This method, however, relies solely on pairwise comparisons, where the generations are evaluated within an identical context. While effective to such conditional preferences often fail to encompass the nuanced and multidimensional nature of human preferences. In this work, we revisit the traditional paradigm of preference acquisition and propose a new axis based on eliciting preferences jointly over the instruction-response pairs. Unlike prior preference optimizations, which are designed for conditional ranking protocols (e.g., DPO), we propose Joint Preference Optimization (JPO), a new preference optimization objective that upweights the joint probability of the chosen instruction-response pair over the rejected instruction-response pair. Interestingly, LLMs trained with joint instruction-response preference data using JPO outperform LLM trained with DPO by 5.2% and 3.3% win-rate for summarization and open-ended dialogue datasets, respectively. Our findings reveal that joint preferences over instruction and response pairs can significantly enhance the alignment of LLMs by tapping into a broader spectrum of human preference elicitation. The data and code is available athttps://github.com/Hritikbansal/jpo.
Probing the Geometry of Truth: Consistency and Generalization of Truth Directions in LLMs Across Logical Transformations and Question Answering Tasks
PDF ↗Large language models (LLMs) are trained on extensive datasets that encapsulate substantial world knowledge. However, their outputs often include confidently stated inaccuracies. Earlier works suggest that LLMs encode truthfulness as a distinct linear feature, termed the “truth direction”, which can classify truthfulness reliably. We address several open questions about the truth direction: (i) whether LLMs universally exhibit consistent truth directions; (ii) whether sophisticated probing techniques are necessary to identify truth directions; and (iii) how the truth direction generalizes across diverse contexts.Our findings reveal that not all LLMs exhibit consistent truth directions, with stronger representations observed in more capable models, particularly in the context of logical negation.Additionally, we demonstrate that truthfulness probes trained on declarative atomic statements can generalize effectively to logical transformations, question-answering tasks, in-context learning, and external knowledge sources.Finally, we explore the practical application of truthfulness probes in selective question-answering, illustrating their potential to improve user trust in LLM outputs.These results advance our understanding of truth directions and provide new insights into the internal representations of LLM beliefs.
Large language models (LLMs) have made significant strides in text-to-SQL tasks; however, small language models (SLMs) are crucial due to their low resource consumption and efficient inference for real-world deployment. Due to resource limitations, SLMs struggle to accurately interpret natural language questions and may overlook critical constraints, leading to challenges such as generating SQL with incorrect logic or incomplete conditions. To address these issues, we propose PARSQL, a novel framework that leverages SQL parsing and reasoning. Specifically, we design PARSer, an SQL parser that extracts constraints from SQL to generate sub-SQLs for data augmentation and producing step-by-step SQL explanations (reason) via both rule-based and LLM-based methods. We define a novel text-to-reason task and incorporate it into multi-task learning, thereby enhancing text-to-SQL performance. Additionally, we employ an efficient SQL selection strategy that conducts direct similarity computation between the generated SQLs and their corresponding reasons to derive the final SQL for post-correction. Extensive experiments show that our PARSQL outperforms models with the same model size on the BIRD and Spider benchmarks. Notably, PARSQL-3B achieves 56.98% execution accuracy on BIRD, rivaling 7B models with significantly fewer parameters, setting a new state-of-the-art performance. Code can be found [here](https://github.com/yaxundai/parsql).
Causal discovery is an imperative foundation for decision-making across domains, such as smart health, AI for drug discovery and AIOps. Traditional statistical causal discovery methods, while well-established, predominantly rely on observational data and often overlook the semantic cues inherent in cause-and-effect relationships. The advent of Large Language Models (LLMs) has ushered in an affordable way of leveraging the semantic cues for knowledge-driven causal discovery, but the development of LLMs for causal discovery lags behind other areas, particularly in the exploration of multi-modal data. To bridge the gap, we introduce MatMCD, a multi-agent system powered by tool-augmented LLMs. MatMCD has two key agents: a Data Augmentation agent that retrieves and processes modality-augmented data, and a Causal Constraint agent that integrates multi-modal data for knowledge-driven reasoning. The proposed design of the inner-workings ensures successful cooperation of the agents. Our empirical study across seven datasets suggests the significant potential of multi-modality enhanced causal discovery.
ConKE: Conceptualization-Augmented Knowledge Editing in Large Language Models for Commonsense Reasoning
PDF ↗Knowledge Editing (KE) aims to adjust a Large Language Model’s (LLM) internal representations and parameters to correct inaccuracies and improve output consistency without incurring the computational expense of re-training the entire model. However, editing commonsense knowledge still faces difficulties, including limited knowledge coverage in existing resources, the infeasibility of annotating labels for an overabundance of commonsense knowledge, and the strict knowledge formats of current editing methods. In this paper, we address these challenges by presenting ConceptEdit, a framework that integrates conceptualization and instantiation into the KE pipeline for LLMs to enhance their commonsense reasoning capabilities. ConceptEdit dynamically diagnoses implausible commonsense knowledge within an LLM using another verifier LLM and augments the source knowledge to be edited with conceptualization for stronger generalizability. Experimental results demonstrate that LLMs enhanced with ConceptEdit successfully generate commonsense knowledge with improved plausibility compared to other baselines and achieve stronger performance across multiple question answering benchmarks. Our data, code, and models are publicly available at https://github.com/HKUST-KnowComp/ConKE.
CMQCIC-Bench: A Chinese Benchmark for Evaluating Large Language Models in Medical Quality Control Indicator Calculation
PDF ↗Medical quality control indicators are essential to assess the qualifications of healthcare institutions for medical services. With the impressive performance of large language models (LLMs) like GPT-4 in the medical field, leveraging these technologies for the Medical Quality Control Indicator Calculation (MQCIC) presents a promising approach. In this work, (1) we introduce a real-world task MQCIC and propose an open-source Chinese electronic medical records (EMRs)-based dataset (CMQCIC-Bench) comprising 785 instances and 76 indicators. (2) We propose a semi-automatic method to enhance the rule representation. Then we propose the Clinical Facts-based Inferential Rule (CF-IR) method that disentangles the clinical fact verification and inferential rule reasoning actions. (3) We conduct comprehensive experiments on 20 representative LLMs, covering general and medical models. Our findings reveal that CF-IR outperforms Chain-of-Thought methods in MQCIC tasks. (4) We conduct an error analysis and investigate the capabilities of clinical fact verification and inferential rule reasoning, providing insights to improve performance in the MQCIC further. The dataset and code is available in this repository https://github.com/YuY-2001/C-MQCIC.
Retrieval-Augmented Language Modeling (RALM) by integrating large language models (LLM) with relevant documents from an external corpus is a proven methodology for enabling the LLM to generate information beyond the scope of its pre-training corpus. Previous work by retrieving a set of tokens iteratively with retrieved content prepending to the input poses a high runtime issue, which degrades the inference efficiency of the LLMs because they fail to use the Key-Value (KV) cache efficiently. We propose FlashBack, a modular RALM designed to improve the inference efficiency of RALM with appending context pattern while maintaining decent performance after fine-tuning by Low-Rank Adaption. FlashBack appends retrieved documents at the end of the context for efficiently utilizing the KV cache. We also introduce the Marking Token as two special prompt tokens for marking the appending context during fine-tuning. Our experiments show that FlashBack can improve language modeling performance in perplexity metric. We proved the Marking Token is a usable add-on when fine-tuning models on specific context patterns. By bypassing unnecessary re-computation, FlashBack achieves fast inference speed speed with long context input. The inference speed is up to 4\times faster than the prepending counterpart on a 7B LLM (Llama 2) in the runtime test.
Analyzing ideological discourse even in the age of LLMs remains a challenge, as these models often struggle to capture the key elements that shape real-world narratives. Specifically, LLMs fail to focus on characteristic elements driving dominant discourses and lack the ability to integrate contextual information required for understanding abstract ideological views. To address these limitations, we propose a framework motivated by the theory of ideological discourse analysis to analyze news articles related to real-world events. Our framework represents the news articles using a relational structure−talking points, which captures the interaction between entities, their roles, and media frames along with a topic of discussion. It then constructs a vocabulary of repeating themes−prominent talking points, that are used to generate ideology-specific viewpoints (or partisan perspectives). We evaluate our framework’s ability to generate these perspectives through automated tasks−ideology and partisan classification tasks, supplemented by human validation. Additionally, we demonstrate straightforward applicability of our framework in creating event snapshots, a visual way of interpreting event discourse. We release resulting dataset and model to the community to support further research.
Large language models (LLMs) have shown great potential in natural language processing tasks, but their application to machine translation (MT) remains challenging due to pretraining on English-centric data and the complexity of reinforcement learning from human feedback (RLHF). Direct Preference Optimization (DPO) has emerged as a simpler and more efficient alternative, but its performance depends heavily on the quality of preference data. To address this, we propose Confidence-Reward driven Preference Optimization (CRPO), a novel method that combines reward scores with model confidence to improve data selection for fine-tuning. CRPO selects challenging sentence pairs where the model is uncertain or underperforms, leading to more effective learning. While primarily designed for LLMs, CRPO also generalizes to encoder-decoder models like NLLB, demonstrating its versatility. Empirical results show that CRPO outperforms existing methods such as RS-DPO, RSO and MBR score in both translation accuracy and data efficiency.
Why Are Positional Encodings Nonessential for Deep Autoregressive Transformers? A Petroglyph Revisited
PDF ↗Do autoregressive Transformer language models require explicit positional encodings (PEs)? The answer is ‘no’ provided they have more than one layer—they can distinguish sequences with permuted tokens without the need for explicit PEs. This follows from the fact that a cascade of (permutation invariant) set processors can collectively exhibit sequence-sensitive behavior in the autoregressive setting. This property has been known since early efforts (contemporary with GPT-2) adopting the Transformer for language modeling. However, this result does not appear to have been well disseminated, leading to recent rediscoveries. This may be partially due to a sudden growth of the language modeling community after the advent of GPT-2/3, but perhaps also due to the lack of a clear explanation in prior work, despite being commonly understood by practitioners in the past. Here we review the long-forgotten explanation why explicit PEs are nonessential for multi-layer autoregressive Transformers (in contrast, one-layer models require PEs to discern order information of their inputs), as well as the origin of this result, and hope to re-establish it as a common knowledge.
We present a simple meta quantization approach that quantizes different layers of a large language model (LLM) at different bit levels, and is independent of the underlying quantization technique. Specifically, we quantize the most important layers to higher bit precision and less important layers to lower bits. We propose two effective strategies to measure the importance of layers within LLMs: the first measures the importance of a layer based on how different its output embeddings are from the input embeddings (higher is better); the second estimates the importance of a layer using the number of layer weights that are much larger than average (smaller is better). We show that quantizing different layers at varying bits as per our importance scores results in minimal performance drop with a far more compressed model. Finally, we present several practical key takeaways from our variable layer-wise quantization experiments: (a) LLM performance under variable quantization remains close to the original model until 25–50% of layers are moved in lower quantization using our proposed ordering but only until 5–10% if moved using no specific ordering; (b) Adding layer importance to inherently dynamic quantization techniques can further improve their performance, showing that our approach is complementary to other dynamic quantization methods; (c) Quantizing LLMs to lower bits performs substantially better than pruning unless extreme quantization (2-bit) is used; and (d) Layer-wise quantization to lower bits works better in the case of larger LLMs with more layers compared to smaller LLMs with fewer layers.
Positional bias in large language models hinders their ability to effectively process long inputs. A prominent example is the “lost in the middle” phenomenon, where LLMs struggle to utilize relevant information situated in the middle of the input. While prior research primarily focuses on single pieces of relevant information, real-world applications often involve multiple relevant information pieces. To bridge this gap, we present LongPiBench, a benchmark designed to assess positional bias involving multiple pieces of relevant information. It includes various tasks and input lengths. Thorough experiments are conducted with three commercial and six open-source models. These experiments reveal that while most current models are more robust against the “lost in the middle” issue, there also exist noticeable biases related to the spacing of relevant information pieces. These findings highlight the importance of evaluating and reducing positional biases for long-context LLMs.