The probabilistic nature of large language models (LLMs) has regularly invited inquiries into their consistency with regard to repeated sampling, input variations, and conceptually related prompts. However, unlike accuracy benchmarks, consistency evaluations cannot rely on standardized measurement tools. We conduct a comprehensive survey of current methodologies evaluating language model consistency across 85 peer-reviewed publications, organizing them into a novel three-level taxonomy: (1) sampling, (2) semantic, and (3) relational consistency. Throughout this process, we identify three fundamental threats to the construct validity of LLM consistency measurements: Valid evaluations must disentangle intrinsic consistency from correctness, probabilistic uncertainty, and artifacts of constrained generation. Our systematic analysis shows that none of the 85 publications accounts for all three factors jointly, allowing construct-irrelevant variance to obscure our understanding of consistency.

Pre-training and fine-tuning large language models (LMs) is currently the state-of-the-art methodology for enabling data-scarce downstream tasks. However, the derived models still tend to replicate and perpetuate social biases. To understand this process in more detail, this paper investigates the actual effects of learning downstream tasks on moral bias in LMs. We develop methods to assess the agreement of LMs to explicitly codified norms in both pre-training and fine-tuning stages. Even if a pre-trained foundation model exhibits consistent norms, we find that introducing downstream tasks may indeed lead to unexpected inconsistencies in norm representation. Specifically, we observe two phenomena during fine-tuning across both masked and causal LMs: (1) pre-existing moral bias may be mitigated or amplified even when presented with opposing views and (2) prompt sensitivity may be negatively impacted. We provide empirical evidence of models deteriorating into conflicting states, where contradictory answers can easily be triggered by slight modifications in the input sequence. Our findings thus raise concerns about the general ability of LMs to mitigate moral biases effectively.

In an effort to better understand the behavior of large language models (LLM), researchers recently turned to conducting psychological assessments on them. Several studies diagnose various psychological concepts in LLMs, such as psychopathological symptoms, personality traits, and intellectual functioning, aiming to unravel their black-box characteristics. But can we safely assess LLMs with tests that were originally designed for humans? The psychology domain looks back on decades of developing standards of appropriate testing procedures to ensure reliable and valid measures. We argue that analogous standardization processes are required for LLM assessments, given their differential functioning as compared to humans. In this paper, we propose seven requirements necessary for testing LLMs. Based on these, we critically reflect a sample of 25 recent machine psychology studies. Our analysis reveals (1) the lack of appropriate methods to assess test reliability and construct validity, (2) the unknown strength of construct-irrelevant influences, such as the contamination of pre-training corpora with test material, and (3) the pervasive issue of non-reproducibility of many studies. The results underscore the lack of a general methodology for the implementation of psychological assessments of LLMs and the need to redefine psychological constructs specifically for large language models rather than adopting them from human psychology.