Almighty ChatGPT? Development of gifted secondary school students’ AI (mis)conceptions.

The current limitations of generative AI (genAI), such as hallucinations, are well-documented.However, research indicates that young students may misconceive genAI as omniscient.How such a misconception develops remains unclear, particularly as the use of advanced AI grows and exposure to AI increases.Building on certain assumptions from the elaboration likelihood model, this study investigates the "omniscient-AI" misconception among gifted secondary school students during a year of rapid AI advancement (2023-2024) by examining its prevalence, persistence, and its association with individual characteristics such as intelligence and personality.A multicohort design yielded crosssectional data across 2 years (N 5 870, Grades 6-8) and longitudinal data for a subsample (n 5 512).Results show that a notable minority (20% in 2023; 16% in 2024) of sixth to eighth graders hold the misconception that genAI is omniscient.Higher intelligence and a greater need for cognition predicted a lower misconception prevalence.Trend analyses suggest a decrease in the prevalence from 2023 to 2024.However, longitudinal analyses revealed no average change in individual students, indicating this trend was driven by a cohort effect rather than individual development, although individual trajectories varied statistically significant.Students with higher intelligence and a higher need for cognition were more likely to revise their conceptions about genAI.The results suggest an emerging Matthew effect, showing that students with a higher capacity and motivation for effortful thinking are better equipped to critically evaluate genAI, potentially exacerbating educational inequalities as those with lower resources remain more susceptible to persistent misconceptions. Educational Impact and Implications StatementThis study finds that a notable number of gifted secondary students believe generative AI (genAI) is all-knowing, a misconception that often persists over time despite increased exposure to AI failures.Students with higher intelligence and a greater enjoyment of thinking are more likely to revise this misconception on their own.These findings show that schools cannot rely on passive experience and must actively teach students how to critically evaluate generative AI's capabilities and limitations.