Researchers have created a new dataset called AraDetox for cleaning up harmful language in Arabic social media posts. The dataset contains 10,500 original posts and their rewritten versions without offensive content, generated using large language models like GPT-5. Human evaluation shows that the rewritten texts successfully remove harmful language while preserving the original meaning. The study also found that detoxification is primarily a rewriting task focused on preserving meaning, rather than just replacing words. --- Why it matters: This matters to researchers in AI because it provides a new resource for studying Arabic text detoxification and developing more effective methods for removing harmful content from social media posts. Source: https://arxiv.org/abs/2608.22894