Lean 4 Programming Language
PulseAugur coverage of Lean 4 Programming Language — every cluster mentioning Lean 4 Programming Language across labs, papers, and developer communities, ranked by signal.
19 day(s) with sentiment data
-
AI agent proves ARC-AGI-3 puzzles unwinnable, outperforming Claude and GPT
Researchers from Tsinghua University and Zhejiang University have developed an Algebraic World Model (AWM) capable of solving ARC-AGI-3 puzzles. This agent, formalized in Lean 4, can also formally prove when a puzzle le…
-
Tsinghua's Algebraic World Model Proves ARC-AGI-3 Levels Unwinnable
Researchers from Tsinghua University and Zhejiang University have developed an Algebraic World Model capable of learning the rules of the ARC-AGI-3 game. This model was then used to formally prove, using the Lean 4 prog…
-
New harness 'h1' matches specialized workflows in formal theorem proving
A new paper introduces 'h1', a simple harness designed to test if scaling test-time compute can match human-designed workflows in formal theorem proving. The harness interfaces with the Lean 4 programming language and u…
-
New DAPS system enhances LLM theorem proving by leveraging dependency signals
Researchers have developed DAPS, a new dependency-aware premise selection system designed to improve Large Language Model (LLM) theorem proving in the Lean 4 programming language. Unlike previous methods that rely on se…
-
Lasm compiles Lean 4 to WebAssembly for verified web apps
Lasm is a new tool that allows developers to write formally verified web applications using the Lean 4 programming language and compile them to WebAssembly. This enables code to be deployed across various JavaScript env…
-
Unified foundation model training algorithm proposed; Van der Waerden number conjecture proven
A new paper proposes a unified training algorithm for foundation models, integrating static data and reinforcement learning into a single likelihood-based approach. This method aims to improve efficiency and capability …
-
New method calibrates AI world model uncertainty for changing policies
Researchers have developed a new method for conformal uncertainty calibration in world models, specifically addressing scenarios where the deployment policy changes while the underlying model remains fixed. This approac…
-
New research quantifies winner's bias in noisy AI evaluations
A new research paper published on arXiv details sharp bounds on the winner's bias in scenarios involving independent and nonnegatively correlated noise. The study quantifies how overstating a winner's value can occur, w…
-
New framework streamlines formal proof generation for math problems
Researchers have developed a novel framework for generating formal proofs in systems like Lean 4, which integrates candidate answer selection and proof generation into a single process. This joint approach aims to reduc…
-
Numina Fuse streamlines AI-assisted math formalization in Lean 4
Numina Fuse is a new agent-native application designed to assist in the formalization of mathematics using the Lean 4 programming language. Developed throughout 2026, it facilitates human supervision of AI agents that w…
-
AI tool KAPPA automates proof-checking in statistics papers
Researchers have developed KAPPA, an open-source software system designed to automatically check proofs in statistical theory papers. KAPPA utilizes language model agents, specifically those powered by Claude Code, to t…
-
New SemantEq method verifies AI-generated formal math statements
Researchers have developed SemantEq, a novel method for verifying the bidirectional equivalence of formal mathematical statements, particularly those generated by AI models. Traditional methods struggle with the increas…
-
New 'h1' system achieves SOTA in automated theorem proving with efficient Lean interface
A new research paper introduces 'h1', a simple agentic loop designed for automated theorem proving. This system utilizes an efficient interface with the Lean 4 programming language and feedback from a verifier to achiev…
-
AI agents now optimize software with machine-checked proofs
Researchers have developed a method called agentic verified optimization, which uses AI agents to improve software performance while ensuring functional equivalence through machine-checked proofs. This approach leverage…
-
PerMIT architecture secures LLM agent memory with integrity theorems
Researchers have developed PerMIT, a novel memory architecture designed to enhance the security and integrity of persistent memory for Large Language Model (LLM) agents. This system employs machine-checked integrity the…
-
Depth Separation in Neural Networks: Depth-3 ReLU Superiority Proven
Researchers have demonstrated that depth-3 ReLU neural networks can approximate certain radial functions with polynomial size in dimensions, while depth-2 networks require exponential size or subexponentially bounded we…
-
New ReuF2F benchmark tests AI coding agents on formal math problems
Researchers have introduced ReuF2F, a new benchmark designed to evaluate AI coding agents on formalized undergraduate research problems. This benchmark consists of 244 open problems drawn from Research Experiences for U…
-
New repair method boosts Lean 4 theorem proving success rate
Researchers have developed a new method for automated theorem proving in Lean 4, a proof assistant. This approach, called bounded compiler-guided repair, improves upon simple sampling models by using compiler error mess…
-
Bidirectionality Proven Necessary for Multi-Agent Path Finding in Cellular Machines
Researchers have demonstrated the necessity of bidirectionality in cellular machines for solving exact-arrival multi-agent path finding problems. This involves studying directed acyclic graphs where each cell must emit …
-
Interactive AI agents improve formal specification accuracy across models
A new research paper explores the effectiveness of interactive clarification agents in improving the accuracy of AI-generated formal specifications. The study found that while a single-shot generation approach yielded i…