Hallucinations in large language models (LLMs), where they generate fluent but factually incorrect outputs, pose challenges for applications requiring strict truthfulness. This work proposes a multi-faceted approach to detect such hallucinations across various language tasks. We leverage automatic data annotation using a proprietary LLM, fine-tuning of the Mistral-7B-instruct-v0.2 model on annotated and benchmark data, role-based and rationale-based prompting strategies, and an ensemble method combining different model outputs through majority voting. This comprehensive framework aims to improve the robustness and reliability of hallucination detection for LLM generations. Code and data1 1 Introduction The modern natural language generation (NLG) (OpenAI et al., 2023; Touvron et al., 2023) landscape faces two interconnected challenges: firstly, current neural models have a tendency to produce f luent yet inaccurate outputs, and secondly, our evaluation metrics are better suited for assessing f luency rather than correctness(Bang et al., 2023; Guerreiro et al., 2023). This phenomenon, known as "hallucination," (Ji et al., 2023) where neural networks generate plausible-sounding but factually incorrect outputs, is a significant hurdle, especially for NLG applications that require strict adherence to correctness. For instance, in machine translation(Lee et al., 2019), producing a fluent translation that deviates from the source text’s meaning renders the entire translation pipeline unreliable. This issue may arise as LLMs are trained on vast amounts of data from the internet, which can contain inaccuracies, biases, and false information. Also, it may arise due improper representations learned during training even if good quality data is 1https://github.com/souvikdgp16/shroom_compos_mentis used. As a result, LLMs can sometimes hallucinate or fabricate details, especially when prompted to discuss topics outside their training data or make inferences beyond their capabilities. Hallucination detection (Liu et al., 2022), also known as factual verification or truthfulness evaluation, identifies and mitigates these hallucinations in the outputs of LLMs. This is an active area of research and development, as it is crucial for ensuring the reliability and trustworthiness of LLMgenerated content, particularly in high-stakes domains such as healthcare, finance, and legal applications. In this task, the primary focus will be to classify whether a generation is hallucinated. This work proposes a multi-faceted approach to detecting hallucinations in large language models.
more »
« less
This content will become publicly available on June 10, 2025
Calibrated Language Models Must Hallucinate
Recent language models generate false but plausible-sounding text with surprising frequency. Such “hallucinations” are an obstacle to the usability of language-based AI systems and can harm people who rely upon their outputs. This work shows that there is an inherent statistical lower-bound on the rate that pretrained language models hallucinate certain types of facts, having nothing to do with the transformer LM architecture or data quality. For “arbitrary” facts whose veracity cannot be determined from the training data, we show that hallucinations must occur at a certain rate for language models that satisfy a statistical calibration condition appropriate for generative language models. Specifically, if the maximum probability of any fact is bounded, we show that the probability of generating a hallucination is close to the fraction of facts that occur exactly once in the training data (a “Good-Turing” estimate), even assuming ideal training data without errors. One conclusion is that models pretrained to be sufficiently good predictors (i.e., calibrated) may require post-training to mitigate hallucinations on the type of arbitrary facts that tend to appear once in the training set. However, our analysis also suggests that there is no statistical reason that pretraining will lead to hallucination on facts that tend to appear more than once in the training data (like references to publications such as articles and books, whose hallucinations have been particularly notable and problematic) or on systematic facts (like arithmetic calculations). Therefore, different architectures and learning algorithms may mitigate these latter types of hallucinations.
more »
« less
- PAR ID:
- 10547047
- Publisher / Repository:
- ACM
- Date Published:
- ISBN:
- 9798400703836
- Page Range / eLocation ID:
- 160 to 171
- Format(s):
- Medium: X
- Location:
- Vancouver BC Canada
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Neural sequence generation models are known to “hallucinate”, by producing outputs that are unrelated to the source text. These hallucinations are potentially harmful, yet it remains unclear in what conditions they arise and how to mitigate their impact. In this work, we first identify internal model symptoms of hallucinations by analyzing the relative token contributions to the generation in contrastive hallucinated vs. non-hallucinated outputs generated via source perturbations. We then show that these symptoms are reliable indicators of natural hallucinations, by using them to design a lightweight hallucination detector which outperforms both model-free baselines and strong classifiers based on quality estimation or large pre-trained models on manually annotated English-Chinese and German-English translation test beds.more » « less
-
Knowledge-dependent tasks typically use two sources of knowledge: parametric, learned at training time, and contextual, given as a passage at inference time. To understand how models use these sources together, we formalize the problem of knowledge conflicts, where the contextual information contradicts the learned information. Analyzing the behaviour of popular models, we measure their over-reliance on memorized information (the cause of hallucinations), and uncover important factors that exacerbate this behaviour. Lastly, we propose a simple method to mitigate over-reliance on parametric knowledge, which minimizes hallucination, and improves out-of-distribution generalization by 4% - 7%. Our findings demonstrate the importance for practitioners to evaluate model tendency to hallucinate rather than read, and show that our mitigation strategy encourages generalization to evolving information (i.e. time-dependent queries). To encourage these practices, we have released our framework for generating knowledge conflicts.more » « less
-
Pre-trained language models (e.g. BART) have shown impressive results when fine-tuned on large summarization datasets. However, little is understood about this fine-tuning process, including what knowledge is retained from pre-training time or how content selection and generation strategies are learnt across iterations. In this work, we analyze the training dynamics for generation models, focusing on summarization. Across different datasets (CNN/DM, XSum, MediaSum) and summary properties, such as abstractiveness and hallucination, we study what the model learns at different stages of its fine-tuning process. We find that a propensity to copy the input is learned early in the training process consistently across all datasets studied. On the other hand, factual errors, such as hallucination of unsupported facts, are learnt in the later stages, though this behavior is more varied across domains. Based on these observations, we explore complementary approaches for modifying training: first, disregarding high-loss tokens that are challenging to learn and second, disregarding low-loss tokens that are learnt very quickly in the latter stages of the training process. We show that these simple training modifications allow us to configure our model to achieve different goals, such as improving factuality or improving abstractiveness.more » « less
-
Language models (LMs) are pretrained on diverse data sources—news, discussion forums, books, online encyclopedias. A significant portion of this data includes facts and opinions 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 media 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 which reinforce the polarization present in pretraining corpora, propagating social biases into hate speech predictions and media biases into misinformation detectors. We discuss the implications of our findings for NLP research and propose future directions to mitigate unfairness.more » « less
