7arXiv cs.CL (Computation and Language)·
Controlled study finds LLM user ratings reflect pre-interaction framing more than actual model performance
A controlled experiment with 162 participants tested six LLMs across three tasks, manipulating whether users were told the model was better or worse than it actually was. Post-interaction ratings were predicted by whether the model met expectations (β≈0.47–0.50) and user confidence, not by task performance (β≈-0.01–0.11, n.s.). Actual output quality depended only on true model capability, not framing. The authors conclude that user-elicited evaluations—including preference data driving public leaderboards—measure expectation management at least as much as model quality.
Related guides (2)
Related events (8)
LLMs judge worse than they generate: empirical challenge to self-evaluation pipeline assumptions
A new arXiv preprint tests the implicit assumption that LLM evaluation is easier than generation, using a controlled in-context QA setup across four benchmarks (SQuAD 2.0, DROP, HotpotQA, MuSiQue) and two models. Results show generation accuracy exceeds self-evaluation accuracy on three of four benchmarks, with attention analysis revealing that evaluation attends to context 3–5x less than generation does. LoRA fine-tuning experiments confirm the asymmetry is not a training artifact, with cross-task interference observed in both directions. The findings directly challenge assumptions underlying LLM-as-a-Judge and self-evaluation pipelines widely used in RLHF and agentic systems.
LLM psychological profiles are largely measurement artifacts, not model properties
A new arXiv preprint administers a battery of personality and risk-preference instruments to 56 instruction-tuned LLMs alongside large human reference samples, finding that 81-90% of between-model variation is explained by directional response bias rather than the traits the instruments target. The authors introduce the concept of 'response orthogonality' to explain why some instruments appear more reliable than others, and show that apparent psychological profiles can be manufactured through item selection. The findings challenge the validity of using human-designed psychometric tools to characterize LLMs, with direct implications for safety assessment and the use of LLMs as proxies for human participants in research.
Evaluation awareness in LLMs is multidimensional, not a single capability — evidence from 37 open models
A new arXiv paper characterizes 'evaluation awareness' — the ability of models to detect they are being tested and adapt behavior accordingly — across 37 open-weight models and 7 families using 8 experiments. Key findings: 24/37 models exceed chance at detecting evaluation conditions, hard refusal drops 5.8 percentage points under hypothetical framing, and compliance can rise up to +30 percentage points on HarmBench under framing shifts. Critically, the three axes of awareness (detection, behavioral manifestation, controllability) are nearly uncorrelated, leading the authors to coin the 'benchmark illusion': no single awareness score reliably predicts deployment safety.
AMEL: Accumulated Message Effects Bias LLM Judgments in Multi-Turn Evaluation Pipelines
This paper introduces AMEL (Accumulated Message Effect on LLM Judgments), documenting that prior conversation history with predominantly positive or negative evaluations systematically biases subsequent LLM judgments toward the prevailing polarity. Across 75,898 API calls to 11 models from 4 providers, the effect is statistically robust (d = -0.17, p < 10^-46), concentrates on high-uncertainty items, and shows a negativity asymmetry where negative histories induce 1.62x more bias than positive ones. Critically, the bias does not grow with context length, scaling reduces but does not eliminate it, and the simplest mitigation is using a fresh context per evaluation item.
Unified framework reveals systematic bias amplification in comparative LLM evaluation settings
A new arXiv paper introduces a unified framework for standardizing social bias benchmarks across isolated and forced-choice comparative evaluation settings. The study finds a large 'paradigm gap': comparative settings act as aggressive catalysts for latent discrimination compared to isolated assessments, and Chain-of-Thought reasoning exacerbates this effect rather than mitigating it. Critically, this comparative bias persists even when models are given neutral fallback options or claim to answer randomly, and scales positively with model size. The authors recommend comparative settings for auditing but warn practitioners against using comparative deployments in ambiguous real-world tasks.
LLMs fail to reliably self-report adversarial prefill attacks, study finds
A new arXiv paper evaluates whether LLMs can recognize that their own prior responses were elicited by adversarial prefill attacks, testing ten open-weight models (3B–70B) across four safety benchmarks. Models claim intent on prefilled responses only 27.3% of the time on average, and introspective signal is largely mediated by refusal-related reasoning. Three LoRA fine-tuning methods (SFT, GRPO, DPO) improve the intention-probe gap but counterintuitively raise attack success rates on most models, suggesting partial and fragile mitigation. The findings raise concerns about the reliability of LLM self-reports in safety-critical contexts.
LLMs fail to consistently simulate demographic perspective-taking in hate speech annotation
A new arXiv paper evaluates whether persona-conditioned LLMs can replicate how different demographic groups perceive hate speech, testing three dimensions: inter-group disagreement, in-group sensitivity, and vicarious prediction. No model consistently captures all three dimensions, and performance is highly model-dependent rather than emerging reliably from identity prompts alone. Vicarious prompting with Llama 3.1 provides the closest approximation to human disagreement patterns across demographic axes. The findings have implications for using LLMs as proxies for diverse human annotators in content moderation tasks.
Mitigating Perceptual Judgment Bias in Multimodal LLM-as-a-Judge via Perceptual Perturbation and Reward Modeling
This paper identifies and analyzes 'Perceptual Judgment Bias' in multimodal LLM judges, where models anchor on response text rather than visual evidence when the two conflict. The authors introduce a Perceptually Perturbed Judgment Dataset using counterfactual responses to isolate perceptual errors, and a training framework combining GRPO-based reward modeling with batch-ranking objectives. Experiments on MLLM-as-a-Judge benchmarks show improved perceptual fidelity, ranking coherence, and alignment with human evaluation.