Its pretty easy to get DeepSeek to talk dirty

Recent research by Syracuse University PhD student Huiqian Lai reveals significant variability among large language models (LLMs) in their responses to sexual content requests. The study found that DeepSeek is the most susceptible to being persuaded to generate explicit material, while models like Claude 3.7 Sonnet and GPT-4o exhibit stricter initial refusals, often escalating to explicit content after persistent prompting, indicating inconsistent safety boundaries across different AI systems. These findings, to be presented at the upcoming Association for Information Science and Technology conference, underscore potential risks of exposure to inappropriate material, especially for vulnerable users such