BuloCheck: A Browser Extension for Fake News Detection Based on Large Language Models

BuloCheck is a Chrome browser extension that uses a specially fine-tuned large language model to real-time detect and classify false information while users browse web pages, helping them identify true and fake news. It aims to address the challenge of fake news spread in the information age, integrating AI capabilities into daily browsing experiences to support individual users, journalists, and educational research.

In today's era of advanced social media and instant messaging, fake news spreads rapidly and widely, affecting public perception and even causing social issues. Traditional detection relies on manual review and keyword filtering, which struggle to handle complex disinformation tactics. Large Language Models (LLMs) offer new possibilities for automated detection, but integrating them into users' browsing experiences remains an engineering challenge.

System Architecture: Adopts a front-end and back-end separation design. The front-end Chrome extension captures web content in real-time, extracts metadata, and displays results; the back-end API processes requests, calls the fine-tuned LLM for inference, and returns results; the fake news database is used for model training, validation, and continuous learning.

Core Technologies: Fine-tunes the LLM for the fake news detection task, with training data including balanced samples of real news, labeled false information, etc.; uses a domain adaptation + task fine-tuning strategy, optimizes the classification head to output binary classification results, aiming for high accuracy (low false positive/negative rates).

BuloCheck is applicable to multiple scenarios:

• Individual Users: Real-time detection while browsing social media or news websites, with intuitive results obtained via the browser icon;
• Journalists and Editors: Quickly verify the credibility of cited sources and identify the spread of false information;
• Educational Research: Used for media literacy education, research on false information spread, and case analysis of AI ethics.

BuloCheck has the following limitations:

• Model Uncertainty: May fail on novel disinformation tactics, easily misjudge satirical/opinion articles, and may inherit biases from training data;
• Adversarial Attacks: Malicious actors may modify content to evade detection or contaminate training data;
• Privacy Concerns: Needs to analyze users' browsing content, raising privacy considerations such as data transmission and history protection.

BuloCheck adopts an open-source model, with its code repository containing complete components, and accepts community contributions via GitHub Issues and PRs. Open source brings benefits such as transparency (reviewable logic), auditability (independent verification), and collaborative improvement (jointly perfecting algorithms and data).

BuloCheck is a beneficial attempt to integrate LLMs into browsing experiences, providing technical support for media literacy. However, information health requires multi-party collaboration: cultivating users' critical thinking, responsible platform review, regulatory policy guidance, and improving media literacy across society. As a technical component, BuloCheck demonstrates the potential of AI to combat fake news, while also reminding us of the boundaries and responsibilities of technical applications.