The paper introduces neuro‑symbolic computer use, a method that learns reusable policies to execute recurring computer workflows efficiently. Instead of re‑planning each run, the learned policy encodes stable decisions (ordering, variables, loops, branches) into executable code while delegating observation‑dependent decisions to neural models. Using neuro‑symbolic policy iteration, the approach iteratively refines the policy from a single agent trajectory, diagnoses failures, and revises the code with a coding model, achieving superior Pass^3 scores and significant reductions in per‑run cost and latency on OSWorld‑Verified and ScienceBoard benchmarks.
By Hyewon Suh, Thanh Minh Nguyen, Chih-Lun Lee, Darrow Hartman, Lizhao Liu, Xin Eric Wang, Ang Li, Jiachen Yang
arXiv:2607. 04096v1 Announce Type: new Abstract: Current agentic workflows usually involve decomposing user requests into sequences of tool calls with correctly resolved parameters, the results of which are processed through reasoning traces in the language model's context window.
By Vishvesh Bhat, Jay Vaghasiya, Emmanuel Anaya Gonzalez
Current agentic workflows usually involve decomposing user requests into sequences of tool calls with correctly resolved parameters, the results of which are processed through reasoning traces in the language model's context window. The prevailing route to improve such reasoning is test-time scaling, which trains models to search over long chains of thought; but the resulting capability is entangled in model weights, is not verifiable step-by-step, and is costly at inference.
arXiv:2604. 05150v2 Announce Type: replace-cross Abstract: We study compiled AI, a paradigm in which large language models generate executable code artifacts during a compilation phase, after which workflows execute deterministically without further model invocation.
By Geert Trooskens (XY.AI Labs, Palo Alto, CA), Aaron Karlsberg (XY.AI Labs, Palo Alto, CA), Anmol Sharma (XY.AI Labs, Palo Alto, CA), Lamara De Brouwer (XY.AI Labs, Palo Alto, CA), Max Van Puyvelde (Stanford University School of Medicine, Stanford, CA), Matthew Young (XY.AI Labs, Palo Alto, CA), John Thickstun (Cornell University, Ithaca, NY), Gil Alterovitz (Brigham and Women's Hospital / Harvard Medical School, Boston, MA), Walter A. De Brouwer (Stanford University School of Medicine, Stanford, CA)
AdaGuard is an adaptive guard model that evaluates language‑model agent trajectories against user‑defined policies. The authors created the AdaptiveSafety dataset, containing 10,939 training and 1,000 test examples with policy rules, counterfactuals, explanations, and violated rules. Using supervised pre‑training followed by the SafePO reinforcement‑learning algorithm, they built 0.6B, 4B, and 8B guard models, with the 4B version achieving 89.30% accuracy on AdaptiveSafety and 71.82% on DynaBench.
By Yunhao Feng, Yifan Ding, Yuxiang Xie, Zheng Li, Mingrui Lao, Zeyuan Wang, Yanming Guo
LangNCO is a framework that transforms natural-language constraints into structured inference-time interventions for a frozen neural combinatorial optimization (NCO) solver. It uses a pretrained large language model to convert each requirement into a domain-specific language rule set, which a deterministic Rule Executor applies as hard action masks. Critical-Offset Decoding then reallocates rollout budget to ensure feasibility, achieving near-perfect feasibility across multiple constraint types without retraining the solver.
By Hironori Ohigashi