Despite their extensive application in language understanding tasks, large language models (LLMs) still encounter challenges including hallucinations - occasional fabrication of information - and alignment issues - lack of associations with human-curated world models (e.g., intuitive physics or common-sense knowledge). Moreover, the black-box nature of LLMs presents significant obstacles in training them effectively to achieve desired behaviors. In particular, modifying the concept embedding spaces of LLMs can be highly intractable. This process involves analyzing the implicit impact of such adjustments on the myriad parameters within LLMs and the resulting inductive biases. We propose a novel architecture that wraps powerful function approximation architectures within an outer, interpretable read-out layer. This read-out layer can be scrutinized to explicitly observe the effects of concept modeling during the training of the LLM. Our method stands in contrast with gradient-based implicit mechanisms, which depend solely on adjustments to the LLM parameters and thus evade scrutiny. By conducting extensive experiments across both generative and discriminative language modeling tasks, we evaluate the capabilities of our proposed architecture relative to state-of-the-art LLMs of similar sizes. Additionally, we offer a qualitative examination of the interpretable read-out layer and visualize the concepts it captures. The results demonstrate the potential of our approach for effectively controlling LLM hallucinations and enhancing the alignment with human expectations.
This content will become publicly available on June 17, 2025
- Award ID(s):
- 2114789
- PAR ID:
- 10502605
- Publisher / Repository:
- Association for Computational Linguistics (ACL)
- Date Published:
- Journal Name:
- North American Chapter of the Association for Computational Linguistics
- Format(s):
- Medium: X
- Location:
- Mexico City
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
-
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
-
more » « less
Building a skilled cybersecurity workforce is paramount to building a safer digital world. However, the diverse skill set, constantly emerging vulnerabilities, and deployment of new cyber threats make learning cybersecurity challenging. Traditional education methods struggle to cope with cybersecurity's rapidly evolving landscape and keep students engaged and motivated. Different studies on students' behaviors show that an interactive mode of education by engaging through a question-answering system or dialoguing is one of the most effective learning methodologies. There is a strong need to create advanced AI-enabled education tools to promote interactive learning in cybersecurity. Unfortunately, there are no publicly available standard question-answer datasets to build such systems for students and novice learners to learn cybersecurity concepts, tools, and techniques. The education course material and online question banks are unstructured and need to be validated and updated by domain experts, which is tedious when done manually. In this paper, we propose CyberGen, a novel unification of large language models (LLMs) and knowledge graphs (KG) to generate the questions and answers for cybersecurity automatically. Augmenting the structured knowledge from knowledge graphs in prompts improves factual reasoning and reduces hallucinations in LLMs. We used the knowledge triples from cybersecurity knowledge graphs (AISecKG) to design prompts for ChatGPT and generate questions and answers using different prompting techniques. Our question-answer dataset, CyberQ, contains around 4k pairs of questions and answers. The domain expert manually evaluated the random samples for consistency and correctness. We train the generative model using the CyberQ dataset for question answering task.
-
UNIWIZ: A Unified Large Language Model Orchestrated Wizard for Safe Knowledge Grounded ConversationsLarge Language Models (LLMs) have made significant progress in integrating safety and knowledge alignment. However, adversarial actors can manipulate these models into generating unsafe responses, and excessive safety alignment can lead to unintended hallucinations. To address these challenges, we introduce UniWiz, a novel 2-step data orchestration framework that unifies safety and knowledge data generation. We propose a “safety-priming” method to generate synthetic safety data and overcome safety bottlenecks. We also inject relevant knowledge into conversations by retrieving factual information from curated sources. UniWiz dataset consists of 17,638 quality-controlled conversations and 10,000 augmented preference data. Pretrained models fine-tuned on UniWiz show improvements across various metrics and outperform state-of-the-art instruction-tuned models trained on much larger datasets.more » « less
-
Despite the significant advancements in the field of Natural Language Processing (NLP), Large Language Models (LLMs) have shown limitations in performing complex tasks that require arithmetic, commonsense, and symbolic reasoning. Reasoning frameworks like ReAct, Chain-of-thought (CoT), Tree-of-thoughts (ToT), etc. have shown success but with limitations in solving long-form complex tasks. To address this, we propose a knowledge-sharing and collaborative multi-agent assisted framework on LLMs that leverages the capabilities of existing reasoning frameworks and the collaborative skills of multi-agent systems (MASs). The objectives of the proposed framework are to overcome the limitations of LLMs, enhance their reasoning capabilities, and improve their performance in complex tasks. It involves generating natural language rationales and in-context few-shot learning via prompting, and integrates the reasoning techniques with efficient knowledge-sharing and communication driven agent networks. The potential benefits of the proposed framework include saving time and money, improved efficiency for computationally intensive reasoning, and the ability to incorporate multiple collaboration strategies for dynamically changing environments.more » « less
