In T382608#11658361, we learned that 1) popular LLMs continue to append metadata to content people copy and paste from them and 2) this "metadata" continues to evolve.
This task involves the work of implementing an LLM-specific version of Paste Check that leverages this metadata.
NOTE: we recognize the instability of this metadata will lead to false negatives (read: the Check not appearing in cases when people paste text directly from LLMs). Even still, the Editing Team thinks the awareness experienced volunteers would gain, and productive friction newcomers would encounter, as a result of this Check seems worthwhile.
Stories
- As someone who is pasting text into a Wikipedia article that I've copied from an LLM in good faith, I want to know what policies/guidelines are relevant to me doing so and what I ought to consider doing in response, so that I can be more confident other volunteers will consider the changes I'm making to be constructive
- As an experienced editor reviewing recent changes, I want to know what edits might contain content pasted from an LLM, so that so that I can A) evaluate the extent to which they might be in violation of LLM-specific Wikipedia policies [1][2][3][4][5][6][7][8] and B) more efficiently review them for issues such as unverifiable claims, hallucinated references, etc.
Open question(s)
- 1. What policy/guideline will each Wikipedia like to include within this Check?
- 2. What options should appear in the decline survey that appears when people elect to Keep the text they're pasting?
- 3.
Requirements
Default configuration
User experience
Check card
- .
- .
- .
Decline survey
- .
- .
- .
Work in Progress
- Visit https://564a50573d.catalyst.wmcloud.org/wiki/Regent's_Park?veaction=edit&ecenable=1 on desktop or mobile
- Copy at least one full paragraph of text from the web interface of ChatGPT, Claude or Gemini
- Paste the text you copied in "2." into the edit session you started in "1."
- ✅ Notice the Potential AI-generated content Check appear
References
- Pangram: strives to detect content generated by AI
- SynthID: watermarks and seeks to identify content generated through AI
- https://en.wikipedia.org/wiki/Wikipedia:Writing_articles_with_large_language_models
- fa:Article creation with large language models
- zh: Writing articles using large language models
- ru: Neuronetwork
- uk: Writer articles using LLM
- az: Use of artificial intelligence
- es: Drafting articles with great language models
- vi: Writing articles using the Big Language Model
