Lean
PulseAugur coverage of Lean — every cluster mentioning Lean across labs, papers, and developer communities, ranked by signal.
5 天有情绪数据
-
Google DeepMind AI solves 9 historic math problems for under $1000
Google DeepMind's AlphaProof Nexus has autonomously solved nine open Erdős mathematical problems, including two that had remained unsolved for 56 years. The AI system, which pairs a large language model with the Lean co…
-
AE Studio uses Modal to train AI for math theorem proving
AE Studio, a consulting partner for Modal, has developed a workflow for training AI models to prove mathematical theorems using reinforcement learning. They compared two methods: Group Relative Policy Optimization (GRPO…
-
AI agents show promise in program verification and theorem proving
Researchers are exploring the use of agentic AI systems, particularly those leveraging large language models (LLMs), for complex tasks like program verification and mathematical theorem proving. Studies show these syste…
-
AI agent solves open math problems using formal proof search
Researchers have developed an AI agent capable of autonomously solving open mathematical problems by generating formal proofs in languages like Lean. This agent successfully resolved 9 out of 353 open Erdős problems and…
-
Algebra and LLMs verify flight-plan bug fix in Lean
Researchers have utilized large language models (LLMs) in conjunction with algebraic methods to verify a bug fix within the Lean theorem prover. This approach focused on a specific flight-plan software component, demons…
-
Lean Refactor optimizes LLM-generated proofs for length and speed
Researchers have developed Lean Refactor, a new framework designed to optimize proofs generated by large language models (LLMs) in the Lean mathematical proof assistant. This system addresses key challenges such as proo…
-
Microsoft Research releases mimalloc high-performance memory allocator
Microsoft Research has released mimalloc, an open-source memory allocator designed for modern, high-concurrency applications and large memory footprints, particularly those involving large language models. This drop-in …
-
MathArena platform evolves to track LLM progress in complex reasoning
Researchers have developed MathArena, an expanded evaluation platform for assessing large language models' mathematical reasoning capabilities. This platform moves beyond static benchmarks to continuously update and bro…
-
GPT-5.4 Pro assists 23-year-old in solving 60-year-old Erdős problem
A 23-year-old individual leveraged GPT-5.4 Pro to solve a 60-year-old mathematical problem known as an Erdős problem. The solution, a proof based on discrete Markov chains, has been verified using the Lean proof assista…
-
AI-generated math proofs lack human insight, hindering understanding
Mathematician David Bessis argues that while AI can generate formal proofs for mathematical theorems, these proofs often lack the explanatory insights crucial for human understanding. He highlights that the process of d…
-
JURY-RL framework enhances LLM reasoning with label-free verifiable rewards
Researchers have developed JURY-RL, a novel framework for label-free reinforcement learning with verifiable rewards (RLVR) designed to improve the reasoning capabilities of large language models. This method separates t…
-
Gemini Deep Think achieves gold-medal standard at International Mathematical Olympiad
An advanced version of Google DeepMind's Gemini model, utilizing its "Deep Think" mode, has achieved a gold-medal standard at the International Mathematical Olympiad (IMO). The model successfully solved five out of six …