Transformer models trained on synthetic pedagogical interaction data in spatial navigation achieve more robust expert-like performance than those trained only on expert demonstrations, particularly when they can distinguish epistemic states of expert and novice agents.
Model predictions were generated using greedy decoding
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Representing expertise accelerates learning from pedagogical interaction data
Transformer models trained on synthetic pedagogical interaction data in spatial navigation achieve more robust expert-like performance than those trained only on expert demonstrations, particularly when they can distinguish epistemic states of expert and novice agents.