Should I move outdoors or ask AI? Influence of setting on performance, interest and self-efficacy in solving Fermi problems

Abstract Fermi problems (FPs) are suitable for introducing real‑world mathematical tasks at the primary level. However, primary school teachers often have difficulties solving FPs and rarely use them in their classrooms. Hence, promoting the successful implementation of FPs in pre-service teachers’ training is necessary. To this end, we aimed to examine whether specific FP-solving settings could facilitate pre-service teachers’ performance, interest and self-efficacy. The study is divided into two phases. First, pre-service teachers solve near-context FPs in one of three different settings: working in the classroom (indoors), working at the actual problem site (outdoors), or working with access to ChatGPT (AI‑support). In the second phase, all pre-service teachers solve more complex FPs with the same AI support. Results indicate that the problem-solving setting significantly influenced outcomes. The outdoor condition led to the highest performance in both phases, with effects persisting into the second phase. Both the outdoor and the AI-supported settings promoted interest, with effects also carrying over to the second phase. Although self‑efficacy was initially lower in the outdoor setting, it showed the greatest improvement later on. These findings suggest that outdoor and AI-supported approaches can benefit pre-service teachers’ own FP-solving performance, interest, and self-efficacy, and may also contribute to preparing them to implement FPs in schools.