paper-with-me

홈 › Papers

Formal Disco: Scalable Open-Ended Generation of Formally Verified Programs

2026-07-06 · Gabriel Poesia, Simon Henniger, Tzu-Han Hsu, Yilun Du, Nada Amin arxiv

The cost of producing code is rapidly diminishing with increasingly capable AI agents, while quality assurance of generated programs has not kept pace. Formal verification provides the strongest possible guarantees, but the ability of AI models to work with verification-aware languages is hindered by the scarcity of human-written examples of programs in those languages. To tackle this prevalent data scarcity issue, we propose Formal Disco: a distributed system for coordination of LLM-based workers that can be easily applied to open-ended synthetic data generation at scale. We use Formal Disco to share tasks and programs between three classes of workers: "initiators", which read random READMEs from open-source repositories and documentation snippets to sketch a related verified program, "fixers" which take compiler and verifier feedback and attempt to resolve issues, and "extenders" that take working programs and propose patches to expand them. Formal Disco records all agent-generated traces and uses them both for initial distillation from a stronger model as well as self-improvement. We also propose a principle of maximum entropy for synthetic program generation, and use entropy maximization via iterative supervised fine-tuning to learn to generate increasingly diverse programs over time. We release large datasets of synthetic verified programs in three languages - Dafny, Verus, and Frama-C -, and fine-tune open models for verification-relevant tasks, often matching or exceeding the performance of Claude Opus 4.5. Overall, our work offers a path to create synthetic data at scale for formal reasoning domains and overcome the long-standing data barrier.

📄 PDF Abstract BibTeX arXiv:2607.04631

Code (1)

Tavish9/awesome-daily-AI-arxiv ★ 111

Tasks

Synthetic Data Generation

Similar Papers 제목 키워드 기반

Bias Association Discovery Framework for Open-Ended LLM Generations

2025-08-02 · Jinhao Pan, Chahat Raj, Ziwei Zhu arxiv

Social biases embedded in Large Language Models (LLMs) raise critical concerns, resulting in representational harms -- unfair or distorted portrayals of demographic groups -- that may be expressed in subtle ways through …

Scalable Formal Concept Analysis algorithm for large datasets using Spark

2018-07-06 · Raghavendra K Chunduri, Aswani Kumar Cherukuri

In the process of knowledge discovery and representation in large datasets using formal concept analysis, complexity plays a major role in identifying all the formal concepts and constructing the concept lattice(digraph …

graph construction

Reverse-Engineered Reasoning for Open-Ended Generation

2025-09-07 · Haozhe Wang, Haoran Que, Qixin Xu, Minghao Liu 외 arxiv

While the ``deep reasoning'' paradigm has spurred significant advances in verifiable domains like mathematics, its application to open-ended, creative generation remains a critical challenge. The two dominant methods for…

Reinforcement Learning

Invariant Discovery for Networked Systems

2026-07-24 · Hongyu Hè, Alexander Krentsel, Sylvia Ratnasamy, Maria Apostolaki arxiv

Invariants, the relations expected to hold among measured signals of a network, underpin applications from verification to traffic generation, telemetry imputation, and input validation, yet writing them by hand demands …

Formal Logic

Open-Endedness is Essential for Artificial Superhuman Intelligence

2024-06-06 · Edward Hughes, Michael Dennis, Jack Parker-Holder, Feryal Behbahani 외

In recent years there has been a tremendous surge in the general capabilities of AI systems, mainly fuelled by training foundation models on internetscale data. Nevertheless, the creation of openended, ever self-improvin…