MALicious INTent Dataset and Inoculating LLMs for Enhanced Disinformation Detection

The project introduces a specific, expert-curated dataset for a nuanced sub-problem (intent vs. truthfulness) in disinformation. While the technical approach (fine-tuning BERT/LLMs) is standard, the data gravity provided by expert fact-checker collaboration offers a small moat. Frontier labs focus more on generative safety than forensic classification, leaving space for this niche.