Reproduce “LLM-IDEA: Identifiability-Driven Experimental Agent for Autonomous Discovery of Mechanistic World Models” 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
Surya Shetty, Ulisses Braga-Neto
LLM-IDEA adds a deterministic diagnostic that tests whether observations identify model parameters and whether permitted experiments can resolve confounded directions. It triages search, redesign within a class, or a change of experiment class. In an audit of 62 ODEBench systems with free constants, it found 60 identifiable, one exhausted for the permitted class, and one resolvable by adding an initial condition; results depend on the observation protocol.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.