With the AI models being trained using Reddit data, do you think by now someone somewhere would have gotten shittymorph’ed?

Exploring AI Exposure to Reddit Content: Have Models Encountered ‘ShittyMorphed’ Styles?

Recent discussions in the AI community have brought to light an intriguing question: Given that large language models are often trained on data sourced from Reddit, is it possible that some models have encountered and learned from niche or meme-specific language styles—particularly those that mimic humorous or intentionally distorted writing, such as the infamous “shittymorph” style?

Motivated by this curiosity, I decided to test whether AI models—specifically Google’s Gemini—are aware of this peculiar Reddit subculture. When prompted to respond in the “shittymorph” style, the model did not disappoint, demonstrating an understanding of the distinctive linguistic twist associated with that community.

This exploration opens up broader questions about the extent of AI’s familiarity with the diverse and often obscure layers of Reddit’s culture. Could delving into more niche subreddits and their unique lexicons help us gauge the depth of a model’s knowledge? Are there other hidden gems within Reddit that AI has been exposed to, which could influence its understanding and responses?

As AI continues to evolve, understanding the scope of its training data is not only fascinating but also essential. It sheds light on the models’ ability to grasp internet slang, meme language, and community-specific styles—elements that shape online communication today.

What are your thoughts? Are there other obscure Reddit traditions or insider language that would be interesting to test or explore further?