Researchers have explored the effectiveness of large language models (LLMs) in detecting phone call scams in Turkish, a low-resource language. They introduced a new dataset of 100 aligned audio-transcript pairs of scam and benign conversations. The study evaluated seven LLMs, including Gemini 2.5 variants, GPT-4o, and Qwen models, using raw audio, automatic transcripts, and human-corrected transcripts. Results indicated that transcript-based inputs were more effective than direct audio processing, with human-corrected and uncorrected transcripts performing similarly. AI
IMPACT Highlights the need for more inclusive AI safety research and multi-modal systems for fraud prevention in low-resource languages.
RANK_REASON The cluster contains an academic paper detailing research on LLM capabilities for a specific task.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
