Large language models produce fluent fiction, yet their creative output is widely seen as flat. We ask where this quality originates in the training and whether it affects different domains of human fiction equally. We construct a matched story-continuation paradigm across StoryStar (public-platform), TMAS (prompt-guided), and The New Yorker (professional literary)—and compare continuations from four OLMo 32B checkpoints (Base, SFT, DPO, RLVR) against matched human text. Because these checkpoints share architecture, scale, tokenizer, and pretraining, the design isolates the post-training effect. We measure each continuation along three sentence-level dimensions: thematic motion, affective prevalence, and linguistic diversity. Across all three, post-training compresses dynamic variation: thematic transitions become more uniform, high-intensity emotions give way to neutrality, and stylistic diversity across stories shrinks. We term this progressive loss narrative flattening. The effect is directionally stable across story domains but gap size depends on the human baseline: professional literary fiction is compressed most, while public-platform and prompt-guided stories show smaller gaps, consistent with their human baselines sitting closer to the model’s default rhythm. Post-trained endpoints converge across domains, suggesting alignment produces a continuation regime largely insensitive to the source domain’s narrative texture.

This position paper argues that researchers should not only study how human biases enter large language models (LLMs), but actively leverage those biases as diagnostic tools to surface society’s "unwritten code"—the tacit norms and hidden heuristics that govern high-stakes decisions yet resist explicit articulation. We present a case study in peer review: a framework that iteratively pushes LLMs to generate self-consistent hypotheses explaining score differences across 26,731 submissions to 46 conferences. LLMs’ normative priors (cues extracted by self-talk) emphasizing theoretical rigor are systematically updated toward posteriors (cues actually used in the judgment) emphasizing storytelling and contextualization—a finding cross-validated by independent human survey evidence. Human reviewers explicitly reward prior-aligned aspects (r=0.49) but rarely articulate the storytelling posteriors (r=-0.14), despite implicitly rewarding them. These findings generalize to out-of-sample judgment and different models. We argue this framework applies to hiring, admissions, and criminal justice, offering a scalable path to making hidden evaluative codes visible for public scrutiny and equitable reform.

Slang is a commonly used type of informal language that poses a daunting challenge to NLP systems. Recent advances in large language models (LLMs), however, have made the problem more approachable. While LLM agents are becoming more widely applied to intermediary tasks such as slang detection and slang interpretation, their generalizability and reliability are heavily dependent on whether these models have captured structural knowledge about slang that align well with human attested slang usages. To answer this question, we contribute a systematic comparison between human and machine-generated slang usages. Our evaluative framework focuses on three core aspects: 1) Characteristics of the usages that reflect systematic biases in how machines perceive slang, 2) Creativity reflected by both lexical coinages and word reuses employed by the slang usages, and 3) Informativeness of the slang usages when used as gold-standard examples for model distillation. By comparing human-attested slang usages from the Online Slang Dictionary (OSD) and slang generated by GPT-4o and Llama-3, we find significant biases in how LLMs perceive slang. Our results suggest that while LLMs have captured significant knowledge about the creative aspects of slang, such knowledge does not align with humans sufficiently to enable LLMs for extrapolative tasks such as linguistic analyses.