LLMs Interpret, Embeddings Organize, Graphs Emerge: Agent-Driven Compilation of Scientific Knowledge
ASKS turns each paper ingest into a traceable update to a persistent wiki-and-graph knowledge base.
The paper defines “scientific knowledge compilation” as preserving interpretation, structure, and source evidence across research tasks. Its system uses an LLM to create readable wiki views and machine-facing semantics, then deterministic checks and graph rules fold those into accumulated state. The authors test it on 56 papers from one research program, producing a source-linked portrait around tensor-network work and related branches. They report stable high-level hub organization, mostly additive canonical-node growth, and graph measurements that retain paths back to source records. ArXiv · AI/CL/LG's note
The paper defines “scientific knowledge compilation” as preserving interpretation, structure, and source evidence across research tasks. Its system uses an LLM to create readable wiki views and machine-facing semantics, then deterministic checks and graph rules fold those into accumulated state. The authors test it on 56 papers from one research program, producing a source-linked portrait around tensor-network work and related branches. They report stable high-level hub organization, mostly additive canonical-node growth, and graph measurements that retain paths back to source records. ArXiv · AI/CL/LG's note
score 4