Evidence for Daily and Weekly Periodic Variability in GPT-4o Performance
This challenges the reliability and reproducibility of research using or investigating LLMs, highlighting a critical issue for the AI research community.
The study tackled the assumption that large language model performance is time-invariant by conducting a longitudinal analysis of GPT-4o, revealing that its average performance varies periodically with daily and weekly rhythms, accounting for about 20% of variance.
Large language models (LLMs) are increasingly used in research both as tools and as objects of investigation. Much of this work implicitly assumes that LLM performance under fixed conditions (identical model snapshot, hyperparameters, and prompt) is time-invariant. If average output quality changes systematically over time, this assumption is violated, threatening the reliability, validity, and reproducibility of findings. To empirically examine this assumption, we conducted a longitudinal study on the temporal variability of GPT-4o's average performance. Using a fixed model snapshot, fixed hyperparameters, and identical prompting, GPT-4o was queried via the API to solve the same multiple-choice physics task every three hours for approximately three months. Ten independent responses were generated at each time point and their scores were averaged. Spectral (Fourier) analysis of the resulting time series revealed notable periodic variability in average model performance, accounting for approximately 20% of the total variance. In particular, the observed periodic patterns are well explained by the interaction of a daily and a weekly rhythm. These findings indicate that, even under controlled conditions, LLM performance may vary periodically over time, calling into question the assumption of time invariance. Implications for ensuring validity and replicability of research that uses or investigates LLMs are discussed.