From Knowledge Access to Source Learning: Developing Source-Specific Competence

Oct 1, 2026ยท
Lucheng Fu
,
Kejing Xia
Yiyang "Diana" Wang
Yiyang "Diana" Wang
,
Yiqiao Jin
,
Jinjin He
,
Xiyuan Yang
,
Haoxin Liu
,
Ye Yu
,
Haibo Jin
,
Yijia Xiao
,
Wenke Lee
,
B. Aditya Prakash
,
Haohan Wang
ยท 1 min read
Type
Publication
arXiv preprint โ€” under submission

SourceLearn studies how LLM agents can progressively learn from a persistent authoritative source. It builds a reusable source model that represents how knowledge is structured, interpreted, and applied.

Self-directed learning revisits incompletely understood material, while task-guided learning uses downstream experience to identify gaps. Updates remain grounded in the authoritative source.