Using neural network language models

Feel free to edit this page!

Just brainstorming.

With the release of ChatGPT for free as a testing phase, language models has shown its potential to either massively improve or disrupt Wikimedia projects. This page is forked from English Wikipedia's w:en:Wikipedia:Using neural network language models on Wikipedia and will focus on the broader Wikimedia stuff.

Don't fret!

It can be said that we've prepared this scenario for years now. While the release of ChatGPT is completely unexpected, we had an in-development project that will be greatly benefited from natural text generation: Abstract Wikipedia. When the project was first announced in 2020, the project aims to make a language-independent version of Wikipedia using Wikidata objects and functions from Wikifunctions, and it has been proposed that Abstract Wikipedia can incorporate natural language text generation (though through a template-like basis). Something that's similar to ChatGPT can greatly reduce the workload needed to translate an article and make the text imported from Abstract Wikipedia far more natural than any predefined algorithm could do.

The WMF has also performed research on how to make ethical and friendly artificial intelligence systems for Wikimedia projects in 2017-2018. And needless to say, we had had incorporated many AI-based tools on Wikimedia projects, as listed in Advanced technology page.