I guess it did partially come true. The GPT-2 model by OpenAI (2019) trained on reddit links did tear some communities apart. :)

https://openai.com/blog/better-language-models/