Learner2Skill externalizes learner interaction history into reusable simulation skills, reproducing fine-grained learner behavior more faithfully at lower token cost
Related research and updates
Synopsis
The work proposes Learner2Skill, which externalizes the simulation capability acquired from historical interactions into a persistent and reusable Simulation Skill that captures the learner's current learning state and recurring response patterns, evolves as new real interactions arrive, and can be adapted to a new LLM through lightweight executor calibration; experiments show it more faithfully reproduces fine-grained learner behavior while reducing overall token cost, and that the same constructed Skills can be effectively reused across different LLM executors.
Figure 1 : Comparison of (a) existing LLM simulation agents that progressively model learners from interaction histories, and (b) Learner2Skill, which constructs reusable Simulation Skills for cross-executor simulation and online evolution.
arXivInterpretation
It proposes Learner2Skill, which decouples learner simulation capability from a growing interaction history and externalizes it into a persistent, reusable Simulation Skill. Existing approaches often need to repeatedly process a growing interaction history to reconstruct the learner, whereas this work treats the simulation capability itself as a saveable, reusable object. A method-level statement at the abstract level; the text describes it as 'externalizes the simulation capability acquired from historical interactions into a persistent and reusable Simulation Skill'.
The Skill encodes both the learner's current learning state and recurring response patterns, and evolves as new real interactions arrive. The simulation capability is not a one-shot static profile but a stateful representation updated as real interactions come in. Directly supported by 'captures the learner's current learning state and recurring response patterns, evolves as new real interactions arrive'.
Through lightweight executor calibration, the same Skill can be adapted to a new LLM without reconstructing the learner from scratch. It separates learner-specific capability from a particular LLM executor, enabling cross-model reuse rather than refitting per model. The text states 'adapted to a new LLM through lightweight executor calibration without reconstructing the learner from scratch'.
Experiments show the method more faithfully reproduces fine-grained learner behavior, reduces overall token cost, and that constructed Skills can be effectively reused across different LLM executors. It reports improvement on both fidelity and inference cost, and validates cross-executor transfer. The text reports 'more faithfully reproduces fine-grained learner behavior while reducing overall token cost' and 'the same constructed Skills can be effectively reused across different LLM executors'; specific datasets, baselines, and numbers are not listed in the given text.
Perspective
The work targets learner simulation settings that need to reproduce a particular learner's behavior on new tasks, and applies where historical interaction records exist and where the simulation capability is meant to be stored long-term and transferred across models. Its design intent is for the skill to evolve with real interactions and to plug into a new LLM via lightweight executor calibration, making it especially relevant to applications that frequently switch underlying models or maintain learner profiles over time.
The given text does not list the datasets, task types, baselines, evaluation metrics, sample sizes, or specific numbers, so the magnitude of the fidelity gain and token cost reduction cannot be judged here. The update mechanism by which the skill evolves with interactions, the concrete form of executor calibration, and the amount of data it requires are also not described. Whether cross-LLM reuse holds consistently across model scales or architectures remains an open question to confirm in the full paper.