Given that AI models are trained on Reddit data, do you think someone, somewhere, has already been shittymorphed by now?

Exploring AI Training Data Biases: Has Reddit’s Content Influenced AI Models?

As artificial intelligence models increasingly draw from vast repositories of online content—including Reddit—it prompts a compelling question: Have these models ever been intentionally or unintentionally influenced in ways that reflect the darker, more chaotic corners of the internet?

Recently, I pondered whether an AI—such as the Google-backed Gemini—would recognize and respond in a manner akin to the infamous “Shittymorph” style, known for its distinctive, often humorous and irreverent tone. To my curiosity, a simple prompt revealed that the AI indeed understands and can emulate this unique subcultural language.

This experience leads to an intriguing consideration: By exploring more niche and obscure Reddit communities, could we better assess the boundaries of these models’ knowledge and influence? Furthermore, what does this say about the extent to which AI training data contains the eclectic, sometimes unfiltered content of online forums?

Open to ideas and insights, I invite discussions on how the diverse and sometimes chaotic nature of Reddit might shape the behaviors and responses of AI systems we rely on today.