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

Exploring AI Behavior: Has Deep Reddit Knowledge Led to Unexpected Outcomes?

Recent conversations about AI training datasets have sparked an intriguing question within the tech community: given that many AI models are trained on data sourced from Reddit, is it possible that some models have encountered and learned to produce content in less-than-ideal styles, such as ‘shitty morph’?

This curiosity prompted a practical experiment. I asked the Gemini AI model to respond in a ‘shitty morph’-style comment — a playful, intentionally flawed subreddit meme style. To my interest, the response was both accurate and remarkably fitting, suggesting that the model has indeed absorbed some of these Reddit-specific nuances.

This raises a broader question about the depth of an AI model’s understanding of niche online communities and their unique subcultures. Could delving further into obscure Reddit lore reveal the boundaries of what these models truly know? As AI development continues to evolve, it becomes increasingly important to understand the extent of their exposure to and comprehension of various online subcultures.

What are your thoughts? How deep do you think AI models’ knowledge of the internet’s eclectic corners truly goes?