The Hyperstition AI-organized fiction contest—the 2026 Unslop AI-Written Fiction Contest—has revealed significant limitations of modern language models in the realm of creative writing.
What Happened
The results of the Unslop AI-Written Fiction Contest have been finalized. The contest featured authors using AI to create fictional texts, with judging conducted by experts including Gwern Branwen, Roon, and Alexander Wales. The primary objective was to produce high-quality ("unslop") prose that is free from typical LLM clichés and repetitive patterns.
Context
The contest aimed to overcome the problem of "stylistic noise" (slop)—the excessive and predictable structures characteristic of standard large language model responses. Participants sought to prove that AI can move beyond simple responses to create texts worthy of attentive reading.
Why It Matters for the Industry
The results demonstrate a clear phenomenon of "mode collapse": even when using complex prompts, models tend to gravitate toward certain stylistic "attractors," such as excessive realism or standardized formats. This creates a technical barrier to generating original content and drives demand for new style control methods, specialized fine-tuning techniques, and the development of architectural solutions (e.g., via RLHF/DPO) optimized for stylistic uniqueness rather than just utility.
Why It Matters for Users
For readers and authors, the contest confirms that current "vanilla" LLMs are not yet capable of deep creativity without significant customization. However, this also opens opportunities for the emergence of new tools and methods for style control that will allow users to bypass model systemic biases toward certain genres and themes.
What Remains Unknown / Limitations
There is a difference in how the scale of the problem is perceived: while technical specialists see mode collapse as a fundamental architectural barrier, business representatives tend to view it as a problem solvable through specialized content management tools.
Sources
Author
Look at AI, Editorial Staff