Following a University of Zurich, Amsterdam, Duke, and NYU research on social media posts on Reddit, X, and Bluesky, it was revealed that AI fails to keep up with human humor, timing, and emotional nuance, highlighting limitations in social media fights.
From ChatGPT writing essays to algorithms generating advertising copy, generative AI’s skill exhibited a range of impressive tasks. Yet, when it comes to AI and online arguments, these intelligent machines can’t fully understand the subtle cognitive dynamics that makes human interaction so unpredictable.
AI Can Copy Form, But Not Feeling
Nine open-weight AI models were tested in the study, including Meta’s Llama, Mistral, and Apertus, with the research revealing that AI generated posts can be differentiated from those created by humans, with around 70 to 80% accuracy.
According to the study, these results suggest that while large language models (LLMs) can reproduce online conversations, they struggle to portray social media sarcasm and feelings.
AI models were generally good at replicating technical features, which exposed some limitations of generative AI social media presence. Posts with sharp humor could almost be attributed to a human, showing the gap between AI and human social skills.
Surprisingly, non-instruction tuned models, such as Llama-3.1-8B and Mistral-7B, outperformed their instruction-tuned versions, suggesting that human-guided alignment can make AI text more uniform and less humanlike.
AI also revealed its weaknesses when responded to heated discussions, showing the challenges in AI handling online fights. These include topics like politics on Reddit or positive comments on X, showing lack of ability in human reasoning.
This reinforces concerns around social media AI escalation, where AI can simplify or misrepresent human disagreement unintentionally.
Automated Social Media Fights
Younger users are abandoning large, centralized platforms, and almost half of Gen Z intend to spend less time on traditional social networks in 2025, reducing their exposure to social media AI arguments.
Short-form video applications, like TikTok and YouTube, expose inadequacies in AI contextual blindness, where machines do not know how to interpret subtle social cues.
Private Discord or WhatsApp conversations give users control over who they interact with, which largely prevents picking fights online. Smaller communities also preserve sarcasm on social media, which AI isn’t quite able to convincingly simulate.
These spaces let human users express humor and emotion, keeping social media nuance and AI in its place. But for all its sophisticated abilities, AI is yet to replace human interaction. Social media fights often sound flat or out of context.
By fostering semi-private, community-driven forums, users create social media without conflict, much like the early internet approach to human-centered engagement. The paper demonstrates that, for the time being, AI can mimic structure but not the rich emotional life that defines authentic online interaction, which keeps social media fights uniquely human.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Tech sections to stay informed and up-to-date with our daily articles.