Reproduce “Forethought: Verifiable Reasoning from Neurosymbolic Primitive Programming” 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
Vishvesh Bhat, Emmanuel Anaya Gonzalez, Jay Vaghasiya, Asad Aali
Forethought makes multi-turn tool-use reasoning explicit by composing specialized neural and symbolic operations in a typed program. It tracks state, orders dependent tasks, binds parameters, and checks each step’s output contract, with failures able to trigger recovery. With Claude Haiku 4.5, its six-benchmark average is 75.9 versus 59.2 for vanilla prompting; results are single runs, and learned steps remain fallible.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.