Can We Count on LLMs? The Fixed-Effect Fallacy and Claims of GPT-4 Capabilities
This highlights a critical methodological flaw in LLM evaluation, showing that claims about capabilities are often overgeneralized and unreliable for researchers and practitioners.
The paper investigates the evaluation of LLM capabilities by measuring GPT-4 performance on deterministic tasks, finding that minor changes in query phrasing or input parameters cause large, statistically significant variations in accuracy, such as differences in list-counting based on composition and frequency.
In this paper we explore evaluation of LLM capabilities. We present measurements of GPT-4 performance on several deterministic tasks; each task involves a basic calculation and takes as input parameter some element drawn from a large well-defined population (e.g., count elements in a list, multiply two k-digit numbers, etc). We examine several conditions per-task and perform enough trials so that statistically significant differences can be detected. This allows us to investigate the sensitivity of task-accuracy both to query phrasing and input parameter population. We find that seemingly trivial modifications in the task-prompt or input population can yield differences far larger than can be explained by sampling effects. For example, performance on a simple list-counting task varies with query-phrasing and list-length, but also with list composition (i.e., the thing-to-be-counted) and object frequency (e.g., success when an element accounts for $\approx$ 50\% of a list is different from when it accounts for $\approx$ 70\% etc). We conclude that efforts to quantify LLM capabilities easily succumb to the language-as-fixed-effect fallacy, where experimental observations are improperly generalized beyond what the data supports. A consequence appears to be that intuitions that have been formed based on interactions with humans form a very unreliable guide as to which input modifications should ``make no difference'' to LLM performance.