Automating victimhood in rage baits

Abstract This article discusses how ChatGPT generates rage bait narratives about migrants in English and Bengali, and situates these synthetic narratives within the prevailing instrumental storytelling culture. Using six different prompt configurations, I examine how the Large Language Model resolves tensions between its internal safety objectives — helpfulness, honesty, and harmlessness — and user instructions. My analysis shows that ChatGPT negotiates competing value objectives through overnarration , whereby narratives redundantly catalog a series of similar but weakly causally connected episodes to establish a character’s innocence or corruption. The type of actions overnarrated in English and Bengali stories varies since the LLM implicitly encourages its users’ identification with the victims of economic and cultural conflicts in each ethnolinguistic context. This narrative structure, I argue, reflects how the LLM’s internalized core principle of “helpfulness” gets operationalized as benefit to an individual user, reinforcing neoliberal ethics, rather than as social responsibility.