Abstract:Large Language Models (LLMs) have significantly advanced communications fields, such as Telecom Q\&A, mathematical modeling, and coding. However, LLMs encounter an inherent issue known as hallucination, i.e., generating fact-conflicting or irrelevant content. This problem critically undermines the applicability of LLMs in communication systems yet has not been systematically explored. Hence, this paper provides a comprehensive review of LLM applications in communications, with a particular emphasis on hallucination mitigation. Specifically, we analyze hallucination causes and summarize hallucination mitigation strategies from both model- and system-based perspectives. Afterward, we review representative LLM-empowered communication schemes, detailing potential hallucination scenarios and comparing the mitigation strategies they adopted. Finally, we present a case study of a Telecom-oriented LLM that utilizes a novel hybrid approach to enhance the hallucination-aware service experience. On the model side, we publish a Telecom hallucination dataset and apply direct preference optimization to fine-tune LLMs, resulting in a 20.6\% correct rate improvement. Moreover, we construct a mobile-edge mixture-of-experts architecture for optimal LLM expert activation. Our research aims to propel the field of LLM-empowered communications forward by detecting and minimizing hallucination impacts.

What problem does this paper attempt to address?

This paper attempts to solve the "hallucination" problem encountered in the application of large - language models (LLMs) in the communication field. Specifically, hallucination refers to the generation of content by LLMs that is inconsistent with facts or irrelevant, which seriously affects the applicability of LLMs in communication systems. Although LLMs have significantly promoted the development of communication technology in aspects such as telecommunication Q&A, mathematical modeling, and coding, the hallucination problem has not been systematically explored yet. To meet this challenge, this article provides a comprehensive review of the application of LLMs in the communication field, with special attention to hallucination mitigation strategies. The main work includes: 1. **Analysis of the Causes of Hallucination**: The article analyzes in detail the possible causes of hallucination in the life cycle of LLMs, including training data bias, model architecture limitations, insufficient domain knowledge, low - quality prompts, and adversarial attacks, etc. 2. **Summary of Hallucination Mitigation Strategies**: - **Model - based Methods**: Such as constructing hallucination detection datasets, using retrieval - enhanced generation (RAG), and performing prompt engineering to improve the performance of LLM. - **System - based Strategies**: Such as federated learning (Federated Learning, FL), mixture of experts (Mixture of Experts, MoE) architecture, secure multi - party computation (Secure Multi - party Computation, SMPC) and other methods to optimize LLM deployment and reduce hallucination. 3. **Evaluation of Existing Schemes**: The article also reviews the existing LLM - enabled communication schemes, evaluates the potential sources of hallucination in these schemes, and compares the mitigation strategies they adopt. 4. **Case Study**: Proposes an LLM for the telecommunications field, using a novel hybrid method for hallucination - aware optimization. Specific measures include releasing a telecommunications - specific hallucination dataset and applying direct preference optimization (Direct Preference Optimization, DPO) to fine - tune the LLM, increasing the correct rate by 20.6%. In addition, a mobile edge MoE architecture is also constructed to optimize the selection of LLM experts, thereby improving the quality of service (QoE) of the entire communication system. In short, this article aims to promote the development of LLM - enabled communication fields by systematically identifying and minimizing the impact of hallucination.

Hallucination-aware Optimization for Large Language Model-empowered Communications

A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models

A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions

Hallucination of Multimodal Large Language Models: A Survey

Hallucination Detection and Hallucination Mitigation: An Investigation

Mitigating Hallucination in Multimodal Large Language Model via Hallucination-targeted Direct Preference Optimization

Quantifying and Attributing the Hallucination of Large Language Models via Association Analysis

SLM Meets LLM: Balancing Latency, Interpretability and Consistency in Hallucination Detection

Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback

The Dawn After the Dark: An Empirical Study on Factuality Hallucination in Large Language Models

Towards Mitigating Hallucination in Large Language Models via Self-Reflection

Alleviating Hallucinations in Large Vision-Language Models through Hallucination-Induced Optimization

Beyond Fine-Tuning: Effective Strategies for Mitigating Hallucinations in Large Language Models for Data Analytics

Cost-Effective Hallucination Detection for LLMs

Hallucination Reduction and Optimization for Large Language Model-Based Autonomous Driving

Mitigating Entity-Level Hallucination in Large Language Models

The Troubling Emergence of Hallucination in Large Language Models -- An Extensive Definition, Quantification, and Prescriptive Remediations

Iter-AHMCL: Alleviate Hallucination for Large Language Model via Iterative Model-level Contrastive Learning

Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models