论文检索

Language · Factuality

Yuzhe Gu, Ziwei Ji, Wenwei Zhang, Chengqi Lyu, Dahua Lin, Kai Chen

Large language models (LLMs) exhibit hallucinations in long-form question-answering tasks across various domains and wide applications. Current hallucination detection and mitigation datasets are limited in domain and size, which struggle to scale due to prohibitive labor costs and insufficient reliability of existing hallucination annotators. To facilitate the scalable oversight of LLM hallucinations, this paper introduces an iterative self-training framework that simultaneously and progressively scales up the annotation dataset and improves the accuracy of the annotator. Based on the Expectation Maximization algorithm, in each iteration, the framework first applies an automatic hallucination annotation pipeline for a scaled dataset and then trains a more accurate annotator on the dataset. This new annotator is adopted in the annotation pipeline for the next iteration. Extensive experimental results demonstrate that the finally obtained hallucination annotator with only 7B parameters surpasses GPT-4 and obtains new state-of-the-art hallucination detection results on HaluEval and HalluQA by zero-shot inference. Such an annotator can not only evaluate the hallucination levels of various LLMs on the large-scale dataset but also help to mitigate the hallucination of LLMs generations, with the Natural Language Inference metric increasing from 25% to 37% on HaluEval.

Qinggang Zhang, Junnan Dong, Hao Chen, Daochen Zha, Zailiang Yu, Xiao Huang

Large Language Models (LLMs) have demonstrated remarkable capabilities in many real-world applications. Nonetheless, LLMs are often criticized for their tendency to produce hallucinations, wherein the models fabricate incorrect statements on tasks beyond their knowledge and perception. To alleviate this issue, graph retrieval-augmented generation (GraphRAG) has been extensively explored which leverages the factual knowledge in knowledge graphs (KGs) to ground the LLM's responses in established facts and principles. However, most state-of-the-art LLMs are closed-source, making it challenging to develop a prompting framework that can efficiently and effectively integrate KGs into LLMs with hard prompts only. Generally, existing KG-enhanced LLMs usually suffer from three critical issues, including huge search space, high API costs, and laborious prompt engineering, that impede their widespread application in practice. To this end, we introduce a novel **Know**ledge **Gr**aph based **P**romp**T**ing framework, namely **KnowGPT**, to enhance LLMs with domain knowledge. KnowGPT contains a knowledge extraction module to extract the most informative knowledge from KGs, and a context-aware prompt construction module to automatically convert extracted knowledge into effective prompts. Experiments on three benchmarks demonstrate that KnowGPT significantly outperforms all competitors. Notably, KnowGPT achieves a 92.6% accuracy on OpenbookQA leaderboard, comparable to human-level performance.

Mingjian Jiang, Yangjun Ruan, Prasanna Sattigeri, Salim Roukos, Tatsunori Hashimoto

Recent advancements in Large Language Models (LLMs) have significantly improved text generation capabilities, but these systems are still known to hallucinate, and granular uncertainty estimation for long-form LLM generations remains challenging. In this work, we propose Graph Uncertainty -- which represents the relationship between LLM generations and claims within them as a bipartite graph and estimates the claim-level uncertainty with a family of graph centrality metrics. Under this view, existing uncertainty estimation methods based on the concept of self-consistency can be viewed as using degree centrality as an uncertainty measure, and we show that more sophisticated alternatives such as closeness centrality provide consistent gains at claim-level uncertainty estimation.Moreover, we present uncertainty-aware decoding techniques that leverage both the graph structure and uncertainty estimates to improve the factuality of LLM generations by preserving only the most reliable claims. Compared to existing methods, our graph-based uncertainty metrics lead to an average of 6.8% relative gains on AUPRC across various long-form generation settings, and our end-to-end system provides consistent 2-4% gains in factuality over existing decoding techniques while significantly improving the informativeness of generated responses.

Interpretability

Samyadeep Basu, Martin Grayson, Cecily Morrison, Besmira Nushi, Soheil Feizi, Daniela Massiceti

Understanding the mechanisms of information storage and transfer in Transformer-based models is important for driving model understanding progress. Recent work has studied these mechanisms for Large Language Models (LLMs), revealing insights on how information is stored in a model's parameters and how information flows to and from these parameters in response to specific prompts. However, these studies have not yet been extended to Multi-modal Large Language Models (MLLMs). Given their expanding capabilities and real-world use, we start by studying one aspect of these models -- how MLLMs process information in a factual visual question answering task. We use a constraint-based formulation which views a visual question as having a set of visual or textual constraints that the model's generated answer must satisfy to be correct (e.g. What movie directed by \emph{the director in this photo} has won a \emph{Golden Globe}?). Under this setting, we contribute i) a method that extends causal information tracing from pure language to the multi-modal setting, and ii) \emph{VQA-Constraints}, a test-bed of 9.7K visual questions annotated with constraints. We use these tools to study two open-source MLLMs, LLaVa and multi-modal Phi-2. Our key findings show that these MLLMs rely on MLP and self-attention blocks in much earlier layers for information storage, compared to LLMs whose mid-layer MLPs are more important. We also show that a consistent small subset of visual tokens output by the vision encoder are responsible for transferring information from the image to these causal blocks. We validate these mechanisms by introducing MultEdit a model-editing algorithm that can correct errors and insert new long-tailed information into MLLMs by targeting these causal blocks. We will publicly release our dataset and code.

Language · Factuality

John Cherian, Isaac Gibbs, Emmanuel Candes

We develop new conformal inference methods for obtaining validity guarantees on the output of large language models (LLMs). Prior work in conformal language modeling identifies a subset of the text that satisfies a high-probability guarantee of correctness. These methods work by filtering claims from the LLM's original response if a scoring function evaluated on the claim fails to exceed a threshold calibrated via split conformal prediction. Existing methods in this area suffer from two deficiencies. First, the guarantee stated is not conditionally valid. The trustworthiness of the filtering step may vary based on the topic of the response. Second, because the scoring function is imperfect, the filtering step can remove many valuable and accurate claims. We address both of these challenges via two new conformal methods. First, we generalize the conditional conformal procedure of Gibbs et al. (2023) in order to adaptively issue weaker guarantees when they are required to preserve the utility of the output. Second, we show how to systematically improve the quality of the scoring function via a novel algorithm for differentiating through the conditional conformal procedure. We demonstrate the efficacy of our approach on biography and medical question-answering datasets.

Language · Factuality

Jianyi Zhang, Da-Cheng Juan, Cyrus Rashtchian, Chun-Sung Ferng, Heinrich Jiang, Yiran Chen

Large language models (LLMs) have demonstrated remarkable capabilities, but their outputs can sometimes be unreliable or factually incorrect. To address this, we introduce Self Logits Evolution Decoding (SLED), a novel decoding framework that enhances the truthfulness of LLMs without relying on external knowledge bases or requiring further fine-tuning. From an optimization perspective, our SLED framework leverages the latent knowledge embedded within the LLM by contrasting the output logits from the final layer with those from early layers. It then utilizes an approximate gradient approach to enable latent knowledge to guide the self-refinement of outputs, thereby effectively improving factual accuracy. Extensive experiments have been conducted on established benchmarks across a diverse range of model families (LLaMA 2, LLaMA 3, Gemma) and scales (from 2B to 70B), including more advanced architectural configurations such as the mixture of experts (MoE). Our evaluation spans a wide variety of tasks, including multi-choice, open-generation, and adaptations to chain-of-thought reasoning tasks. The results demonstrate that SLED consistently improves factual accuracy by up to 20\% compared to existing decoding methods while maintaining natural language fluency and negligible latency overhead. Furthermore, it can be flexibly combined with other decoding methods to further enhance their performance.

Weichao Zhao, Hao Feng, Qi Liu, Jingqun Tang, Binghong Wu, Lei Liao, Shu Wei, Yongjie Ye, Hao Liu, Wengang Zhou 等

Tables contain factual and quantitative data accompanied by various structures and contents that pose challenges for machine comprehension. Previous methods generally design task-specific architectures and objectives for individual tasks, resulting in modal isolation and intricate workflows. In this paper, we present a novel large vision-language model, TabPedia, equipped with a concept synergy mechanism. In this mechanism, all the involved diverse visual table understanding (VTU) tasks and multi-source visual embeddings are abstracted as concepts. This unified framework allows TabPedia to seamlessly integrate VTU tasks, such as table detection, table structure recognition, table querying, and table question answering, by leveraging the capabilities of large language models (LLMs). Moreover, the concept synergy mechanism enables table perception-related and comprehension-related tasks to work in harmony, as they can effectively leverage the needed clues from the corresponding source perception embeddings. Furthermore, to better evaluate the VTU task in real-world scenarios, we establish a new and comprehensive table VQA benchmark, ComTQA, featuring approximately 9,000 QA pairs. Extensive quantitative and qualitative experiments on both table perception and comprehension tasks, conducted across various public benchmarks, validate the effectiveness of our TabPedia. The superior performance further confirms the feasibility of using LLMs for understanding visual tables when all concepts work in synergy. The benchmark ComTQA has been open-sourced at https://huggingface.co/datasets/ByteDance/ComTQA. The source code and model also have been released at https://github.com/zhaowc-ustc/TabPedia.

Trustworthy Machine Learning

Yuxin Xiao, Wan Chaoqun, Yonggang Zhang, Wenxiao Wang, Binbin Lin, Xiaofei He, Xu Shen, Jieping Ye

As the development and application of Large Language Models (LLMs) continue to advance rapidly, enhancing their trustworthiness and aligning them with human preferences has become a critical area of research. Traditional methods rely heavily on extensive data for Reinforcement Learning from Human Feedback (RLHF), but representation engineering offers a new, training-free approach. This technique leverages semantic features to control the representation of LLM's intermediate hidden states, enabling the model to meet specific requirements such as increased honesty or heightened safety awareness. However, a significant challenge arises when attempting to fulfill multiple requirements simultaneously. It proves difficult to encode various semantic contents, like honesty and safety, into a singular semantic feature, restricting its practicality.In this work, we address this challenge through Sparse Activation Control. By delving into the intrinsic mechanisms of LLMs, we manage to identify and pinpoint modules that are closely related to specific tasks within the model, i.e. attention heads. These heads display sparse characteristics that allow for near-independent control over different tasks. Our experiments, conducted on the open-source Llama series models, have yielded encouraging results. The models were able to align with human preferences on issues of safety, factualness, and bias concurrently.

Deep Learning · Attention Mechanisms

Yibo Jiang, Goutham Rajendran, Pradeep Ravikumar, Bryon Aragam

Large Language Models (LLMs) have the capacity to store and recall facts. Through experimentation with open-source models, we observe that this ability to retrieve facts can be easily manipulated by changing contexts, even without altering their factual meanings. These findings highlight that LLMs might behave like an associative memory model where certain tokens in the contexts serve as clues to retrieving facts. We mathematically explore this property by studying how transformers, the building blocks of LLMs, can complete such memory tasks. We study a simple latent concept association problem with a one-layer transformer and we show theoretically and empirically that the transformer gathers information using self-attention and uses the value matrix for associative memory.

Language · Knowledge

Hoyeon Chang, Jinho Park, Seonghyeon Ye, Sohee Yang, Youngkyung Seo, Du-Seong Chang, Minjoon Seo

Despite the recent observation that large language models (LLMs) can store substantial factual knowledge, there is a limited understanding of the mechanisms of how they acquire factual knowledge through pretraining. This work addresses this gap by studying how LLMs acquire factual knowledge during pretraining. The findings reveal several important insights into the dynamics of factual knowledge acquisition during pretraining. First, counterintuitively, we observe that pretraining on more data shows no significant improvement in the model's capability to acquire and maintain factual knowledge. Next, LLMs undergo forgetting of memorization and generalization of factual knowledge, and LLMs trained with duplicated training data exhibit faster forgetting. Third, training LLMs with larger batch sizes can enhance the models' robustness to forgetting. Overall, our observations suggest that factual knowledge acquisition in LLM pretraining occurs by progressively increasing the probability of factual knowledge presented in the pretraining data at each step. However, this increase is diluted by subsequent forgetting. Based on this interpretation, we demonstrate that we can provide plausible explanations on recently observed behaviors of LLMs, such as the poor performance of LLMs on long-tail knowledge and the benefits of deduplicating the pretraining corpus.

Language · Factuality

Zhihua Wen, Zhiliang Tian, Zexin Jian, Zhen Huang, Pei Ke, Yifu Gao, Minlie Huang, Dongsheng Li

Large Language Models (LLMs) are widely used for knowledge-seeking purposes yet suffer from hallucinations. The knowledge boundary of an LLM limits its factual understanding, beyond which it may begin to hallucinate. Investigating the perception of LLMs' knowledge boundary is crucial for detecting hallucinations and LLMs' reliable generation. Current studies perceive LLMs' knowledge boundary on questions with concrete answers (close-ended questions) while paying limited attention to semi-open-ended questions that correspond to many potential answers. Some researchers achieve it by judging whether the question is answerable or not. However, this paradigm is not so suitable for semi-open-ended questions, which are usually ``partially answerable questions'' containing both answerable answers and ambiguous (unanswerable) answers. Ambiguous answers are essential for knowledge-seeking, but it may go beyond the knowledge boundary of LLMs. In this paper, we perceive the LLMs' knowledge boundary with semi-open-ended questions by discovering more ambiguous answers. First, we apply an LLM-based approach to construct semi-open-ended questions and obtain answers from a target LLM. Unfortunately, the output probabilities of mainstream black-box LLMs are inaccessible to sample more low-probability ambiguous answers. Therefore, we apply an open-sourced auxiliary model to explore ambiguous answers for the target LLM. We calculate the nearest semantic representation for existing answers to estimate their probabilities, with which we reduce the generation probability of high-probability existing answers to achieve a more effective generation. Finally, we compare the results from the RAG-based evaluation and LLM self-evaluation to categorize four types of ambiguous answers that are beyond the knowledge boundary of the target LLM. Following our method, we construct a dataset to perceive the knowledge boundary for GPT-4. We find that GPT-4 performs poorly on semi-open-ended questions and is often unaware of its knowledge boundary. Besides, our auxiliary model, LLaMA-2-13B, is effective in discovering many ambiguous answers, including correct answers neglected by GPT-4 and delusive wrong answers GPT-4 struggles to identify.

Language · Generation

Ethan Shen, Alan Fan, Sarah Pratt, Jae Sung Park, Matthew Wallingford, Sham Kakade, Ari Holtzman, Ranjay Krishna, Ali Farhadi, Aditya Kusupati

Many applications today provide users with multiple auto-complete drafts as they type, including GitHub's code completion, Gmail's smart compose, and Apple's messaging auto-suggestions. Under the hood, language models support this by running an autoregressive inference pass to provide a draft. Consequently, providing $k$ drafts to the user requires running an expensive language model $k$ times. To alleviate the computation cost of running $k$ inference passes, we propose Superposed Decoding, a new decoding algorithm that generates $k$ drafts at the computation cost of one autoregressive inference pass. We achieve this by feeding a superposition of the most recent token embeddings from the $k$ drafts as input to the next decoding step of the language model. At every inference step we combine the $k$ drafts with the top-$k$ tokens to get $k^2$ new drafts and cache the $k$ most likely options, using an n-gram interpolation with minimal compute overhead to filter out incoherent generations. Our experiments show that $k$ drafts from Superposed Decoding are at least as coherent and factual as Nucleus Sampling and Greedy Decoding respectively, while being at least $2.44\times$ faster for $k\ge3$. In a compute-normalized setting, user evaluations demonstrably favor text generated by Superposed Decoding over Nucleus Sampling. Superposed Decoding can also be combined with other decoding strategies, resulting in universal coverage gains when scaling inference time compute. Code and more examples open-sourced at https://github.com/RAIVNLab/SuperposedDecoding.

Trustworthy Machine Learning

Alexander Nikitin, Jannik Kossen, Yarin Gal, Pekka Marttinen

Uncertainty quantification in Large Language Models (LLMs) is crucial for applications where safety and reliability are important. In particular, uncertainty can be used to improve the trustworthiness of LLMs by detecting factually incorrect model responses, commonly called hallucinations. Critically, one should seek to capture the model's semantic uncertainty, i.e., the uncertainty over the meanings of LLM outputs, rather than uncertainty over lexical or syntactic variations that do not affect answer correctness.To address this problem, we propose Kernel Language Entropy (KLE), a novel method for uncertainty estimation in white- and black-box LLMs. KLE defines positive semidefinite unit trace kernels to encode the semantic similarities of LLM outputs and quantifies uncertainty using the von Neumann entropy. It considers pairwise semantic dependencies between answers (or semantic clusters), providing more fine-grained uncertainty estimates than previous methods based on hard clustering of answers. We theoretically prove that KLE generalizes the previous state-of-the-art method called semantic entropy and empirically demonstrate that it improves uncertainty quantification performance across multiple natural language generation datasets and LLM architectures.

Interpretability

Lennart Bürger, Fred Hamprecht, Boaz Nadler

Large Language Models (LLMs) have revolutionised natural language processing, exhibiting impressive human-like capabilities. In particular, LLMs are capable of "lying", knowingly outputting false statements. Hence, it is of interest and importance to develop methods to detect when LLMs lie. Indeed, several authors trained classifiers to detect LLM lies based on their internal model activations. However, other researchers showed that these classifiers may fail to generalise, for example to negated statements. In this work, we aim to develop a robust method to detect when an LLM is lying. To this end, we make the following key contributions: (i) We demonstrate the existence of a two-dimensional subspace, along which the activation vectors of true and false statements can be separated. Notably, this finding is universal and holds for various LLMs, including Gemma-7B, LLaMA2-13B, Mistral-7B and LLaMA3-8B. Our analysis explains the generalisation failures observed in previous studies and sets the stage for more robust lie detection;(ii) Building upon (i), we construct an accurate LLM lie detector. Empirically, our proposed classifier achieves state-of-the-art performance, attaining 94\% accuracy in both distinguishing true from false factual statements and detecting lies generated in real-world scenarios.

Language · Factuality

Gaurang Sriramanan, Siddhant Bharti, Vinu Sankar Sadasivan, Shoumik Saha, Priyatham Kattakinda, Soheil Feizi

While Large Language Models (LLMs) have become immensely popular due to their outstanding performance on a broad range of tasks, these models are prone to producing hallucinations— outputs that are fallacious or fabricated yet often appear plausible or tenable at a glance. In this paper, we conduct a comprehensive investigation into the nature of hallucinations within LLMs and furthermore explore effective techniques for detecting such inaccuracies in various real-world settings. Prior approaches to detect hallucinations in LLM outputs, such as consistency checks or retrieval-based methods, typically assume access to multiple model responses or large databases. These techniques, however, tend to be computationally expensive in practice, thereby limiting their applicability to real-time analysis. In contrast, in this work, we seek to identify hallucinations within a single response in both white-box and black-box settings by analyzing the internal hidden states, attention maps, and output prediction probabilities of an auxiliary LLM. In addition, we also study hallucination detection in scenarios where ground-truth references are also available, such as in the setting of Retrieval-Augmented Generation (RAG). We demonstrate that the proposed detection methods are extremely compute-efficient, with speedups of up to 45x and 450x over other baselines, while achieving significant improvements in detection performance over diverse datasets.

Language · Knowledge

Roi Cohen, Konstantin Dobler, Eden Biran, Gerard de Melo

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known as hallucinations, causing them to emit unwanted and factually incorrect text. In this work, we propose a novel calibration method that can be used to combat hallucinations. We add a special [IDK] (“I Don't Know”) token to the model's vocabulary and introduce an objective function that shifts probability mass to the [IDK] token for incorrect predictions. This approach allows the model to express uncertainty in its output explicitly. We evaluate our proposed method across multiple model architectures and factual downstream tasks.We find that models trained with our method are able to express uncertainty in places where they would previously make mistakes while suffering only a small loss of encoded knowledge. We further perform extensive ablation studies of multiple variations of our approach and provide a detailed analysis of the precision-recall tradeoff of our method.

Language · Factuality

Jerry Wei, Chengrun Yang, Xinying Song, Yifeng Lu, Nathan Hu, Jie Huang, Dustin Tran, Daiyi Peng, Ruibo Liu, Da Huang 等

Large language models (LLMs) often generate content that contains factual errors when responding to fact-seeking prompts on open-ended topics. To benchmark a model’s long-form factuality in open domains, we first use GPT-4 to generate LongFact, a prompt set comprising thousands of questions spanning 38 topics. We then propose that LLM agents can be used as automated evaluators for long-form factuality through a method which we call Search-Augmented Factuality Evaluator (SAFE). SAFE utilizes an LLM to break down a long-form response into a set of individual facts and to evaluate the accuracy of each fact using a multi-step reasoning process comprising sending search queries to Google Search and determining whether a fact is supported by the search results. Furthermore, we propose extending F1 score as an aggregated metric for long-form factuality. To do so, we balance the percentage of supported facts in a response (precision) with the percentage of provided facts relative to a hyperparameter representing a user’s preferred response length (recall).Empirically, we demonstrate that LLM agents can outperform crowdsourced human annotators—on a set of∼16k individual facts, SAFE agrees with crowdsourced human annotators 72% of the time, and on a random subset of 100 disagreement cases, SAFE wins 76% of the time. At the same time, SAFE is more than 20 times cheaper than human annotators. We also benchmark thirteen language models on LongFact across four model families (Gemini, GPT, Claude, and PaLM-2), finding that larger language models generally achieve better long-form factuality. LongFact, SAFE, and all experimental code are available at https://github.com/google-deepmind/long-form-factuality.

Language · Factuality

Yifu QIU, Zheng Zhao, Yftah Ziser, Anna Korhonen, Edoardo Maria Ponti, Shay Cohen

Large language models (LLMs) often exhibit undesirable behaviours, such as generating untruthful or biased content. Editing their internal representations has been shown to be effective in mitigating such behaviours on top of the existing alignment methods. We propose a novel inference-time editing method, namely spectral editing of activations (SEA), to project the input representations into directions with maximal covariance with the positive demonstrations (e.g., truthful) while minimising covariance with the negative demonstrations (e.g., hallucinated). We also extend our method to non-linear editing using feature functions. We run extensive experiments on benchmarks concerning truthfulness and bias with six open-source LLMs of different sizes and model families. The results demonstrate the superiority of SEA in effectiveness, generalisation to similar tasks, as well as computation and data efficiency. We also show that SEA editing only has a limited negative impact on other model capabilities.

Language · Factuality

Hyunseok Lee, Jihoon Tack, Jinwoo Shin

The remarkable capabilities and easy accessibility of large language models (LLMs) have significantly increased societal risks (e.g., fake news generation), necessitating the development of LLM-generated text (LGT) detection methods for safe usage. However, detecting LGTs is challenging due to the vast number of LLMs, making it impractical to account for each LLM individually; hence, it is crucial to identify the common characteristics shared by these models. In this paper, we draw attention to a common feature of recent powerful LLMs, namely the alignment training, i.e., training LLMs to generate human-preferable texts. Our key finding is that as these aligned LLMs are trained to maximize the human preferences, they generate texts with higher estimated preferences even than human-written texts; thus, such texts are easily detected by using the reward model (i.e., an LLM trained to model human preference distribution). Based on this finding, we propose two training schemes to further improve the detection ability of the reward model, namely (i) continual preference fine-tuning to make reward model prefer aligned LGTs even further and (ii) reward modeling of Human/LLM mixed texts (a rephrased texts from human-written texts using aligned LLMs), which serves as a median preference text corpus between LGTs and human-written texts to learn the decision boundary better. We provide an extensive evaluation by considering six text domains across twelve aligned LLMs, where our method demonstrates state-of-the-art results.

Interpretability

Maximilian Li, Lucas Janson

Interpretability studies often involve tracing the flow of information through machine learning models to identify specific model components that perform relevant computations for tasks of interest. Prior work quantifies the importance of a model component on a particular task by measuring the impact of performing ablation on that component, or simulating model inference with the component disabled. We propose a new method, optimal ablation (OA), and show that OA-based component importance has theoretical and empirical advantages over measuring importance via other ablation methods. We also show that OA-based component importance can benefit several downstream interpretability tasks, including circuit discovery, localization of factual recall, and latent prediction.