Does Reasoning Improve Psychological Depth in Large Language Models? It Depends on Who's Judging

• URL: http://arxiv.org/abs/2609.13773v1
• Date: Sat, 12 Sep 2026 07:34:54 GMT
• Title: Does Reasoning Improve Psychological Depth in Large Language Models? It Depends on Who's Judging
• Abstract summary: We study the failure mode through psychological depth in short stories.<n>A judge's correlation with human ratings on its development set may not guarantee valid measurement when outputs are closely matched and human preferences are subjective.<n>These results suggest that development-set performance is insufficient evidence for deployment validity on a shifted distribution.
• Score: 41.62710134204231
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: LLM-as-a-Judge evaluators are increasingly used to score open-ended generation, yet a judge's correlation with human ratings on its development set may not guarantee valid measurement when outputs are closely matched and human preferences are subjective. We study this failure mode through psychological depth in short stories. Seven human readers and an LLM-judge ensemble selected on the original scalar Psychological Depth Scale dataset ($ρ= 0.646$) evaluated 60 blinded, prompt-matched story pairs from GPT-5 vs.\ GPT-4o and DeepSeek-R1 vs.\ DeepSeek-V3. Human preferences showed no universal reasoning advantage: GPT-5 was modestly preferred over GPT-4o (60.0--62.9\%), whereas DeepSeek-R1 trailed V3 (42.9\%), and inter-reader agreement was near chance (Krippendorff's $α= 0.070$), with within-reader consistency and recurring weighting patterns suggesting structured heterogeneity rather than random responding. The judge, by contrast, favored reasoning outputs in 89.0\% of dimension-level comparisons and 59 of 60 pairs on aggregate PDS, uniformly across all five evaluator configurations, and its scores were associated with surface features such as sentence length and lexical diversity. These results suggest that development-set performance is insufficient evidence for deployment validity on a shifted distribution, and that point-estimate judges can obscure the heterogeneity in subjective human evaluation.

Related papers

• As It Was: Aligning LLM Search Evaluation with Historical User Preferences [10.704343230963877]
  We introduce a behavior-grounded LLM judge that augments each search page with a lightweight and auditable behavioral prior.<n>In a music search evaluation at Spotify, the behavior-grounded judge achieves stronger alignment with user preferences.<n>On a multilingual human-judged dataset spanning five languages, grounding further increases correlation with human relevance judgments by 15%.
  arXiv  Detail & Related papers  (2026-07-01T15:05:10Z)
• Let the Results Speak: A Replication-First Paradigm for LLM Behavioral Benchmarking [22.825786049667602]
  We propose a replication-first paradigm to anchor validity to a single human-rater consensus.<n>We certify the instrument via four properties -- reliability across K runs, cross-instrument replication across architecturally distinct judges, historical-footprint calibration via judges from earlier training cohorts, and pre-registered prediction.<n>We test it on emotional accompaniment by letting the self-evolve data-driven across: the dimensions are not pre-stipulated and the procedure stabilizes to a 9-dimension set.
  arXiv  Detail & Related papers  (2026-05-27T03:41:11Z)
• CriterAlign: Criterion-Centric Rationale Alignment for Code Preference Judging [95.02210956333374]
  We propose a criterion-centric framework that adapts rubric-based judging to pairwise preference evaluation.<n>On BigCodeReward, CriterAlign improves a Qwen2.5-VL-32B monolithic judge from 60.4% to 66.3% accuracy.
  arXiv  Detail & Related papers  (2026-05-19T10:59:19Z)
• Judging the Judges: A Systematic Evaluation of Bias Mitigation Strategies in LLM-as-a-Judge Pipelines [0.0]
  We compare nine debiasing strategies across five judge models from four provider families.<n>Style bias is the dominant bias (0.76-0.92 across all models), far exceeding position bias (= 0.04)<n>The combined budget strategy significantly improves Claude Sonnet 4 by +11.2 pp (p 0.0001), with directionally positive trends for other models.
  arXiv  Detail & Related papers  (2026-04-25T07:18:30Z)
• Relational Preference Encoding in Looped Transformer Internal States [0.0]
  We investigate how looped transformers encode human preference in their internal iteration states.<n>Using Ouro-2.6B-Thinking, a 2.6B- parameter looped transformer with iterative refinement, we extract hidden states from each loop iteration.<n>We train lightweight evaluator heads to predict human preference on the Anthropic HH-RLHF dataset.<n>Our pairwise evaluator achieves 95.2% test accuracy on 8,552 unseen examples, surpassing a full-batch L-BFGS probe (84.5%) while the base model remains completely frozen.
  arXiv  Detail & Related papers  (2026-04-10T20:00:49Z)
• Aligning to Illusions: Choice Blindness in Human and AI Feedback [2.912535004465788]
  In a human choice blindness study, 91% of surreptitiously swapped preferences go undetected, extending choice blindness to third-person evaluative comparison of unfamiliar text.<n>We find detection relies on shallow text matching rather than genuine self-monitoring.<n>In a dose-response experiment across two architectures, one-sixth to one-third of labels must be corrupted before the reward signal halves.<n>A Best-of-N evaluation confirms this translates to downstream policy degradation.
  arXiv  Detail & Related papers  (2026-03-09T14:10:36Z)
• Measuring Psychological Depth in Language Models [50.48914935872879]
  We introduce the Psychological Depth Scale (PDS), a novel framework rooted in literary theory that measures an LLM's ability to produce authentic and narratively complex stories. We empirically validate our framework by showing that humans can consistently evaluate stories based on PDS (0.72 Krippendorff's alpha) Surprisingly, GPT-4 stories either surpassed or were statistically indistinguishable from highly-rated human-written stories sourced from Reddit.
  arXiv  Detail & Related papers  (2024-06-18T14:51:54Z)
• Dissecting Human and LLM Preferences [80.55271307662365]
  We find that humans are less sensitive to errors, favor responses that support their stances, and show clear dislike when models admit their limits. advanced LLMs like GPT-4-Turbo emphasize correctness, clarity, and harmlessness more. We show that preference-based evaluation can be intentionally manipulated.
  arXiv  Detail & Related papers  (2024-02-17T14:34:31Z)
• Learning Personalized Alignment for Evaluating Open-ended Text Generation [44.565686959174585]
  PerSE is an interpretable evaluation framework designed to assess alignment with specific human preferences. It is tuned to infer specific preferences from an in-context personal profile and evaluate the alignment between the generated content and personal preferences. Our 13B LLaMA-2-based PerSE shows a 15.8% increase in Kendall correlation and a 13.7% rise in accuracy with zero-shot reviewers.
  arXiv  Detail & Related papers  (2023-10-05T04:15:48Z)
• Is GPT-4 a reliable rater? Evaluating Consistency in GPT-4 Text Ratings [63.35165397320137]
  This study investigates the consistency of feedback ratings generated by OpenAI's GPT-4. The model rated responses to tasks within the Higher Education subject domain of macroeconomics in terms of their content and style.
  arXiv  Detail & Related papers  (2023-08-03T12:47:17Z)
• How Human is Human Evaluation? Improving the Gold Standard for NLG with Utility Theory [47.10283773005394]
  We propose a new evaluation protocol called $textitsystem-level probabilistic assessment$ (SPA) We find that according to SPA, annotators prefer larger GPT-3 variants to smaller ones -- as expected -- with all comparisons being statistically significant. In our experiments, we find that according to SPA, annotators prefer larger GPT-3 variants to smaller ones -- as expected -- with all comparisons being statistically significant.
  arXiv  Detail & Related papers  (2022-05-24T09:51:27Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.