Dafny
PulseAugur coverage of Dafny — every cluster mentioning Dafny across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
LLM code generation fails on verification, new paper finds
A new research paper, "Verified but Vacuous: Three Failure Surfaces of LLM-Driven Verified Code Generation," published on arXiv, argues that current metrics for evaluating LLM-generated code are insufficient. The paper …
-
AI agents write code that passes verifiers but has incomplete specifications
An experiment using AI coding agents to generate formally verified code revealed that while models like Claude Code could produce solutions that passed verifiers, the specifications themselves were often incomplete. The…
-
New auditing method doubles Verus/Rust program correctness evidence discovery
A new research paper on arXiv introduces a method called Phase-Separated Completeness Auditing for Verus/Rust programs. This technique aims to improve the discovery of program correctness by avoiding feedback loops that…
-
New benchmark 'claimcheck' tackles AI-generated specification faithfulness
Researchers have developed a new benchmark called claimcheck to measure and improve the faithfulness of formal specifications generated by AI models. The benchmark covers four formal languages—Dafny, Lean 4, Isabelle, a…
-
SkillForge framework generates formally verified Dafny programs
Researchers have developed SkillForge, a novel framework designed to generate formally verified Dafny programs from natural language descriptions. This system decomposes the complex task into a library of reusable skill…
-
New method Re:Form uses formal languages to reduce human annotation for LLM software verification
Researchers have developed a new method called Re:Form to reduce the need for human annotations in training Large Language Models (LLMs) for formal software verification. By leveraging formal languages like Dafny and in…
-
New system Formal Disco generates verified code datasets at scale
Researchers have developed Formal Disco, a scalable system designed to generate large datasets of formally verified programs. This system uses a distributed approach with three types of AI workers: initiators to sketch …
-
New benchmark MINIF2F-DAFNY tests LLMs for mathematical theorem proving
Researchers have developed MINIF2F-DAFNY, a new benchmark for evaluating Large Language Models (LLMs) in mathematical theorem proving. This system translates the miniF2F benchmark to Dafny, an auto-active verifier, enab…
-
New benchmark reveals AI struggles with verified code generation
A new benchmark called AlgoVeri has been developed to evaluate the performance of AI models in generating formally verified code for classical algorithms. The benchmark tests models across three languages: Dafny, Verus,…
-
AI models improve code generation with new verification techniques
Researchers have developed new methods to improve the ability of large language models to generate correct code and proofs. One approach, TTRL-CoCoV, uses confidence-conditioned verification to enhance coverage and accu…
-
Researchers develop graph construction for imperative programs using neural methods
Researchers have developed a pipeline to convert imperative programs and their annotations into typed, attributed graphs. This process combines abstract syntax tree parsing with semantic embeddings from models like Sent…
-
SEVerA framework verifies self-evolving AI agents for safety and correctness
Researchers have introduced SEVerA, a framework designed to synthesize self-evolving AI agents with formal safety and correctness guarantees. This approach treats agentic code generation as a constrained learning proble…
-
AI models achieve high verification success with formal code generation
Researchers have developed a new dataset, NL2VC-60, containing 60 algorithmic problems to aid in generating verified code from natural language. They evaluated seven open-weight LLMs using various prompting strategies, …