Preprints
https://doi.org/10.5194/egusphere-2026-6339
https://doi.org/10.5194/egusphere-2026-6339
15 Oct 2026
 | 15 Oct 2026
Status: this preprint is open for discussion.

Listen. Answer. Memorize. Building and Running Verified Learning Paths for Geoscience Self-Study with an AI Assistant

Stefan Mohr

Abstract. AI tutors built on large language models can state things that no source supports, a risk that weighs on self-study. This paper describes how a single adult learner built, with an AI assistant, learning paths from his own verified fact documents in the geosciences and, for two of the five cases, learned them in voice calls. The skill lernpfad-sitzungen cuts a fact document into sessions of whole paragraphs with three questions each, whose expected answers must appear word for word in the excerpt of the session; a check script verifies this. In a session, the assistant reads the excerpt aloud, asks the questions, grades the spoken answers tolerantly and keeps a learning record, from which review sessions are built on request. The study covers five cases: four learning projects, on southern Norway, Earth’s geological eras, the geology of Germany and volcanology, and a narrative version of the fact document on the geological eras. All six learning paths checked, among them two versions of the path on southern Norway, passed the check script, three of them only after chapters that are not learning material had been named as exclusions. By 13 October 2026, the learning records held 25 sessions on two cases, 23 of them entirely in a voice call, and two review sessions had been completed. By the learner’s own account, sessions of five to ten minutes let him learn on the move and in short breaks. Incidents in building learning paths and in sessions produced new rules for the assistant; in the narrative version, for example, all questions were asked as worded only once they were shown in the text chat before each call. The study is a self-experiment without a control condition and makes no claim about retention. The rules of the skill contain no subject content; in the author’s assessment, it should also suit other subjects that rely on factual and conceptual knowledge, which has not been tested.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
Share
Stefan Mohr

Status: open

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
Stefan Mohr

Data sets

Github Repository: Listen. Answer. Memorize. Stefan Mohr https://github.com/nordie69/Listen-Answer-Memorize

Stefan Mohr
Metrics will be available soon.
Latest update: 15 Oct 2026
Download
Short summary
This self-experiment examines an AI-assisted workflow for geoscience self-study using verified fact documents. A script checks that learning paths cover the material and that expected answers appear verbatim in each lesson. Across 25 sessions in two cases, the assistant read passages aloud, graded spoken answers and maintained learning records. The learner reported that 5–10-minute sessions supported learning on the move. Retention was not measured.
Share