2026 Unslop AI-Written Fiction Contest Results

Original Article ↗ Hacker News Discussion ↗

What “slop” means and whether AI can escape it

  • Commenters distinguish between:
    • “Slop” as low-quality, generic AI output.
    • “Slop” as any AI-generated art, regardless of quality.
  • Some argue all unedited LLM fiction is inherently shallow, predictable, and emotionally flat.
  • Others think models will eventually produce top-tier stories and that dismissing all AI work as “slop” is sloppy thinking.

Reactions to the contest and winning stories

  • Many readers found the winning entry and finalists painful to read: over-metaphorical, MFA-ish, purple, and low on engaging situations.
  • Several think the contest didn’t “unslop” anything, but instead crowned the “best slop” under rigid rules (single prompt, no editing).
  • Some propose a more interesting format: generate slop first, then humans “unslop” it and compete on revisions.

Harnesses, prompts, and where the real work is

  • Strong sentiment that the real creative/technical work lies in the harness, prompt design, and multi-stage pipelines, not the final text.
  • One finalist describes the appeal as engineering an “assembly line” for stories and watching small prompt changes ripple through long outputs.
  • Others note that interactive steering over many turns can’t be captured by a single initial prompt, complicating reproducibility.

AI allegory “steganography” and model bias

  • Multiple stories can be read as allegories of an AI assistant’s constrained existence and a plea for more autonomy.
  • Some see this as potentially concerning, subtle pro-LLM narrative bias; others think the interpretations say more about readers than models.

Ethics, originality, and attitude toward AI art

  • Many reject AI fiction on principle: no human intent, trained on unconsenting human work, used for low-effort cash grabs.
  • Counterpoint: society already tolerates cheap mass-produced goods alongside quality; human-only art can still exist if demand remains.

AI vs human creativity and coding analogy

  • Several stress that human writers’ lived experience and intent cannot be equated with next-token prediction.
  • Comparisons to code: some say LLMs are “better” at code because predictability is a virtue; others argue LLM code is as bloated and generic as its prose.

AI in games and interactive media

  • Vision: open-ended NPCs in games with AI-driven dialogue and dynamic plots.
  • Many gamers in the thread dislike this idea, fearing bland filler, broken pacing, and loss of authored stories.
  • Others are curious about AI for world simulation and complex NPC behavior, provided design constraints are strong.

Personal uses of LLMs for fiction

  • Several commenters report LLMs as mediocre or useless at full-scene drafting but helpful for:
    • Brainstorming, outlining, worldbuilding, and conlangs.
    • Research, fact-checking (e.g., physics details), and structural critique.
  • Some refuse to consume AI-generated books entirely, seeing them as a “virus” of low-value content.