Social media’s popularity has contributed to an increase in hate speech, endangering people’s safety and fueling hate crimes. Spelling mistakes, code-mixing, and a variety of dialects make it challenging to detect such material, particularly in low-resource languages with few high-quality datasets. Using long language models, the current study investigates six prompting tactics for hate speech detection in Bengali, including a unique metaphor prompting technique that gets over common LLM safety filters. All methods are tested on the Llama2-7B model and compared to deep learning models and three word embeddings. The researchers also use F1 score and environmental impact indicators to evaluate performance on Hindi, English, and German in order to demonstrate generalizability. https://arxiv.org/abs/2506.23930 Share this: Click to share on Facebook (Opens in new window) Facebook Click to share on X (Opens in new window) X Like this:Like Loading... Post navigation Social Hatred: Efficient Multimodal Detection of Hatemongers (arXiv)