Year: 2026
Type: conference-paper
Abstract: What Has Quietly Changed? Creation is cheap Verification is slow Responsibility is unclear This gap is where trust breaks.• No Restriction Models: C4AI, Zephyr-orpo, Mistral: Generate fake news easily without safety protocols.• Moderate Restriction Models: Phi-3, Llama-3: Some hesitation but still generate fake news in specific contexts.• Strict Adherence Models: GPT-4, Gemma-1.1:Refuse fake news generation due to robust safety measures. Key InsightLarger models and better-trained systems show more restraint.Can LLMs Generate Fake News? ModelPhi-3Can we use LLMs to Detect Fake News?• Human-Created Fake News Detection Most models struggled to provide definitive classifications for real vs.fake news.• Top Performers: Llama-3 & Zephyr-orpo • Underperformers: Mistral • Notable Performance: GPT-4: High accuracy with no errors in correct classifications.However, frequent inconclusive outcomes present challenges for user decision-making.Can we use LLMs to Detect Fake News?• LLM-Generated Fake News Detection ▪ Larger models outperform smaller ones in accuracy.▪ GPT-4 excels but returns inconclusive results due to cautiousness.• Human-in-the-loop verification: Designing systems where AI supports detection, but humans remain accountable for decisions• Context-aware misinformation detection: Moving beyond binary true/false labels to include intent, impact, and social context• Explainable AI for trust: Making detection and moderation decisions transparent and understandable to users Research & Future Directions• AI literacy and assessment frameworks: Embedding augmented intelligence principles into education and evaluation systems• Policy-aligned AI systems: Aligning NLP tools with emerging regulations and ethical standards Research & Future Directions• The future of AI is not about smarter machines.• It is about wiser human decisions.• Augmented intelligence is how we protect truth, trust, and accountability.
Author: Kumar, Akshi
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7163721574