Reproduce “Lineage Is Memory: Parent Inheritance and Feedback Disclosure in Short-Budget LLM-Guided Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shanglin Yang, Zhan Shi
The authors test whether LLM optimizers should restart each revision from an original prompt or build on the best candidate so far. Across seven primary tasks, parent inheritance raises held-out accuracy by 0.238 over seed-reset across four disclosure schemes. Even with full-failure summaries, parent-first leads by 0.157. However, primary tasks favor accumulating withheld rules; extension-task effects are smaller and inconclusive.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.