Show HN: AiSniff, underline the phrases in a text that read like AI wrote them (github.com)

AiSniff, a new tool, allows users to identify phrases in their text that match common writing patterns associated with artificial intelligence. It marks these phrases in both English and Chinese, providing a rule name, public source, and suggested plain rewrites for clarity. Operating entirely from a set of JSON rules without server calls, AiSniff offers functionality through a web editor, command line interface (CLI), and as an agent skill. Its utility lies in the ability to highlight phrases that may read like AI-generated content, which can help users refine their writing and avoid unintentional AI-like patterns. This tool is significant for the AI and machine learning community as it addresses the growing concern over distinguishing AI-produced text from human writing. The evaluation of over 23,000 human and chatbot responses revealed that about 29% of chatbot answers and only 2% of human answers contained AI-like phrases in English. Although the density of detected phrases offers weak detection capability (AUROC of 0.63 for English), AiSniff's design is meant to aid in enhancing textual clarity rather than confirming authorship. By providing a structured approach to identifying AI writing patterns, AiSniff fosters greater awareness around the use and implications of AI-generated content in various contexts.