tech
Wikipedia volunteers spent years cataloging AI tells. Now there's a plugin to avoid them.
The web’s best guide to spotting AI writing has become a manual for hiding it.
TL;DR
- Siqi Chen released an open-source plugin called "Humanizer" for Anthropic's Claude Code AI assistant.
- The plugin instructs Claude to stop writing like an AI by using a list of 24 language and formatting patterns identified by Wikipedia editors.
- The source material for the patterns is a guide from WikiProject AI Cleanup, a group of Wikipedia editors.
- The Humanizer tool is a "skill file" for Claude Code, which appends instructions to the prompt fed into the LLM.
- Limited testing indicates the plugin can make AI output sound less precise and more casual, but may have drawbacks like not improving factuality or potentially harming coding ability.
- Some instructions in the Humanizer, like encouraging opinions, might be detrimental for tasks like technical documentation.
- Wikipedia editors identified patterns such as inflated language and phrases like "marking a pivotal moment" as chatbot giveaways.
- The article notes that AI writing detectors are not always reliable, as humans can write in chatbot-like ways, and LLMs can be prompted to avoid certain language patterns.