The new tool aims to leverage Wikipedia’s AI detection guide to help AI chatbots generate more human-like text. ars technica. Developer Siqi Chen said he created the tool, called Humanizer, by providing Anthropic’s Claude with a list of information compiled by Wikipedia’s volunteer editors as part of an effort to combat “AI-generated, poorly written content.”
Wikipedia’s guide includes a list of signs that the text may have been generated by AI. This includes things like vague attribution, blurbs that describe something as “breathtaking,” and supportive phrases like “I hope this helps!” According to its GitHub page, Claude Code’s custom skill, Humanizer, can help evade detection by AI assistants by removing these “AI-generated writing signatures” from text, making it more natural and human-like.
The GitHub page provides examples of how Humanizer can help Claude discover some of this information, such as changing the text that a location is “surrounded by breathtaking territory” to “a town in the region of Gondar,” or adjusting a vague attribution like “experts believe it plays an important role” to “according to a 2019 study by…” Chen said the tool will “automatically push updates” when Wikipedia’s AI detection guide is updated. Updated.
OpenAI has already addressed the overuse of em dashes in ChatGPT as indicators of AI content, so it’s only a matter of time before AI companies themselves start adjusting their chatbots to some of these instructions.
