Reddit users, curious and restless as they await more updates regarding FromSoftware’s forthcoming Switch 2 exclusive, *The Duskbloods*, have begun crafting their own stories about the game. Much of this has been expressed through a trend known as “Duskposting,” where imaginative users invent details about non-existent bosses, offer descriptions, and even create strategy guides to assist in overcoming these fictional adversaries. As highlighted by Ian Walker in *Video Game Town*, this movement started soon after the game’s announcement, reflecting a blend of community involvement and playful creativity. Nevertheless, it presents a distinctive challenge, particularly regarding how large language models (LLMs) like ChatGPT and others interpret and share information.
A major weakness of LLMs is their vulnerability to misinformation, mainly because they serve as sophisticated predictive text generators rather than genuine intelligences. They do not grasp the significance of the content they produce, which results in a notable susceptibility to selectively absorbing and recycling unvalidated information. This can create amusing yet troubling situations, where entirely invented details are viewed as truthful.
A notable example of this was illustrated by BBC reporter Thomas Germain, who penned a satirical blog entry claiming that tech reporters had participated in a fictional hot dog-eating competition. Within hours, this made-up account was being cited by AI models, illustrating their ability to confidently share misinformation derived from less reliable sources.
As excitement for *The Duskbloods* grows, it seems Reddit users have influenced LLMs similarly. For instance, a straightforward search query like “Which bosses are toughest to defeat in *The Duskbloods*?” prompted Gemini, an AI model, to propose names such as “Lycan Vrykolakas,” “Dawn-Eater Abomination,” and “Jocasta, Eminence of the Night.” None of these bosses, nor their descriptions, have been validated by FromSoftware as part of the game’s narrative. Instead, the AI’s suggestions were sourced from discussions on the Duskbloods subreddit, demonstrating how it recycled imaginative content as credible information.
The AI ventured deeper into fictive strategy territory by analyzing the “most challenging boss,” the Dawn-Eater Abomination, detailing mechanics like mob spawns and confined arena spaces as primary challenges. This speculative guide was solely based on Reddit discussions, devoid of any actual gameplay insight or authentic game development details.
Even when asked to provide strategies for specific encounters—like surviving the second phase against the Dawn-Eater Abomination—the AI generated an entire strategy replete with detailed suggestions as though it were genuine advice from an experienced player. This scenario accentuates not only the entertainment value of creative Reddit contributions but also the potential risks associated with relying on AI-generated information that is not factually grounded.
The growing reliance on AI for information regrettably exacerbates the skepticism that can stem from receiving unverifiable guidance. A candid dialogue with the AI, wherein it was informed that it had shared incorrect data based on a singular, unreliable source, resulted in an amusing yet unsettling admission from the model: “The Duskbloods is not a real video game.”
This demonstrates both the capacity for community-driven creativity and the necessity for caution, highlighting the importance of validating sources amid the rise of AI-generated narratives. As gaming culture continues to progress with interactive and creative input from users, so too must our interactions with emerging technologies that process and deliver this content. The Duskbloods remains a much-anticipated subject, with users eager to explore its creative vein, but the misrepresentation by AI could result in confusion in a realm where reality and fiction converge.
