**FACT-GPT: Fact-Checking Augmentation via Claim Matching with LLMs** The paper introduces FACT-GPT, a system that leverages Large Language Models (LLMs) to automate the claim matching stage in fact-checking. This stage involves identifying social media content that aligns with, contradicts, or is irrelevant to previously debunked claims. The system is trained on a synthetic dataset, and its performance is evaluated against human judgment, showing that specialized LLMs can match the accuracy of larger models in identifying related claims. **Introduction:** The rapid proliferation of misinformation on digital platforms has highlighted the need for efficient fact-checking. Claim matching, a critical component of this process, involves identifying new instances of previously fact-checked claims. Effective claim matching can expedite the detection of misinformation, content moderation, and automated debunking. **Related Work:** AI models that support fact-checkers are more likely to be accepted, as they enhance efficiency without replacing human expertise. LLMs have shown promise in various annotation tasks, but their probabilistic nature and task-specific limitations must be considered. **Proposed Framework:** - **Task Description:** The framework uses a textual entailment task to categorize relationships between pairs of statements into Entailment, Neutral, and Contradiction. - **Datasets:** The study focuses on misinformation related to public health, specifically COVID-19 false claims. Synthetic training data is generated using LLMs, and a ground truth dataset is created through human annotation. - **Experiments:** Pre-trained LLMs are compared against human annotations, and fine-tuning is performed on synthetic datasets. The results show that fine-tuned models, particularly those trained with GPT-4-generated data, perform better in identifying relevant or irrelevant content but struggle with detecting contradictory posts. **Discussion:** The study demonstrates the potential of LLMs in augmenting fact-checking, especially in claim matching. While smaller LLMs can achieve comparable performance to larger models, further research is needed to improve the detection of contradictory posts. Ongoing collaborations among researchers, developers, and fact-checkers are crucial to fully realize the benefits of AI while mitigating potential drawbacks. **Conclusion:** This research adds to the growing body of work on using LLMs to support human fact-checkers, offering a foundation for future studies and responsible advancements in AI tools to combat misinformation.**FACT-GPT: Fact-Checking Augmentation via Claim Matching with LLMs** The paper introduces FACT-GPT, a system that leverages Large Language Models (LLMs) to automate the claim matching stage in fact-checking. This stage involves identifying social media content that aligns with, contradicts, or is irrelevant to previously debunked claims. The system is trained on a synthetic dataset, and its performance is evaluated against human judgment, showing that specialized LLMs can match the accuracy of larger models in identifying related claims. **Introduction:** The rapid proliferation of misinformation on digital platforms has highlighted the need for efficient fact-checking. Claim matching, a critical component of this process, involves identifying new instances of previously fact-checked claims. Effective claim matching can expedite the detection of misinformation, content moderation, and automated debunking. **Related Work:** AI models that support fact-checkers are more likely to be accepted, as they enhance efficiency without replacing human expertise. LLMs have shown promise in various annotation tasks, but their probabilistic nature and task-specific limitations must be considered. **Proposed Framework:** - **Task Description:** The framework uses a textual entailment task to categorize relationships between pairs of statements into Entailment, Neutral, and Contradiction. - **Datasets:** The study focuses on misinformation related to public health, specifically COVID-19 false claims. Synthetic training data is generated using LLMs, and a ground truth dataset is created through human annotation. - **Experiments:** Pre-trained LLMs are compared against human annotations, and fine-tuning is performed on synthetic datasets. The results show that fine-tuned models, particularly those trained with GPT-4-generated data, perform better in identifying relevant or irrelevant content but struggle with detecting contradictory posts. **Discussion:** The study demonstrates the potential of LLMs in augmenting fact-checking, especially in claim matching. While smaller LLMs can achieve comparable performance to larger models, further research is needed to improve the detection of contradictory posts. Ongoing collaborations among researchers, developers, and fact-checkers are crucial to fully realize the benefits of AI while mitigating potential drawbacks. **Conclusion:** This research adds to the growing body of work on using LLMs to support human fact-checkers, offering a foundation for future studies and responsible advancements in AI tools to combat misinformation.