Can a System-One LLM Perform Knowledge Tracing When Few or No Learners Are Logged?

arXiv:2610.11135v1 Announce Type: cross Abstract: Knowledge tracing (KT) models need many logged learners, so a new course or platform starts without a usable model. In LLM-based KT the LLM generates the answer, which we call System-Two; it is either fine-tuned on the target data or reasons and votes over ten samples, which is slow and gives coarse probabilities. We ask whether an off-the-shelf System-One LLM, which returns a probability for a typed question directly in a single pass, can perform KT when few or no learners are logged. On seven datasets, Jev without any data from the target platform reaches a mean AUC of .706, above the best of 28 deep KT models trained on 8 learners (.689) and above System-Two Thinking-KT on all seven datasets (.650) at about 1/100 of its API cost. Adding examples and a similar-learner statistic from the logged learners (JevKT) raises this to .722; JevKT stays significantly ahead of deep KT up to 16 learners and ahead on average up to 64, and supervised KT catches up between 64 and 128 learners. Among the readers we tested, the gain is specific to Jev, since three other LLMs queried with the byte-identical typed request through the official System-One adapter fall below it on all seven datasets, and reader swaps and contamination checks find no evidence that the input format or memorised data explain the gain. For new learners the advantage holds from their first interactions, whereas on unseen items with all learners logged, deep KT remains ahead.

Jev LLM shows promise in knowledge tracing without learner data.

This is a narrow technical benchmarking study on LLM-based knowledge tracing with no direct claims about broad societal flourishing. It has mild positive implications for cheaper, scalable educational tools but stays within a specific academic niche.

This paper presents a significant improvement in AI's ability to perform knowledge tracing, particularly in new educational contexts with limited data. The described method achieves higher accuracy and lower cost than existing deep learning models and other LLMs, enabling more effective personalized learning experiences from the outset. This represents a substantial positive development for educational technology.

The study demonstrates a significant improvement in knowledge tracing using a System-One LLM. This advancement could lead to more efficient and effective educational tools, particularly in new courses or platforms with few logged learners.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   cb075f91-b3e6-45c9-a512-48b034fbe1b4
           https://arxiv.org/abs/2610.11135
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  4 runs, methodology 1.0.1
   ↓
SCORE      -26  (Adverse)
   ↓
CONFIDENCE 75%