Pre-train and fine-tune large language models, and combine them with reasoning algorithms.
Build and maintain efficient machine learning pipelines.
Collaborate with other teams to understand their priorities and define roadmap milestones.
Summarised from the posting by AI and can be wrong. The full description is the source.
Requirements
Experience
5+ years of subject matter expertise and research in Machine Learning, Deep Learning, or Reasoning
Experience with explicit and implicit reasoning
Demonstrated research publications at major conferences such as CVPR, ICLR, ICML, NeurIPS, ICCV, AAAI, or ACL
Research expertise in Energy-based Modeling (preferred)
A proven record of using Energy-based Modeling for different problems (preferred)
Hands-on experience with algorithms used to train Energy-based models (preferred)
Education
M.Sc. or Ph.D. focused on Computer Science, Artificial Intelligence, Mathematics, or a closely related field (preferred)
Skills
Expertise in Python and C++ for high-performance computing
Expertise with machine learning frameworks including PyTorch, TensorFlow, or JAX
Multi-node and multi-GPU training (preferred)
Mathematical reasoning, including discrete math and logic (preferred)
Formal verification with Lean (preferred)
Full job description
From Logical Intelligence’s careers site
Who we are
Show full descriptionShow less
At Logical Intelligence, we're revolutionizing software development with AI-powered formal verification. We've developed groundbreaking agents that provide mathematical guarantees of code correctness, ensuring that software behaves exactly as intended while proactively identifying bugs and security vulnerabilities. Our novel foundation model enables scalable, precise reasoning for formally verifiable code across Rust, Golang, and smart contract VMs. We’ve won a well-known formal verification benchmark called PutnamBench, which consists of 672 hard math problems from the William Lowell Putnam Exam, the oldest collegiate mathematics competition in North America. Backed by a world-class team – including ICPC champions, a Fields Medalist and an ACM Turing Award winner – we're building the future where all code is provably correct.
About the role
Join our team as an AI Researcher and help us push the boundaries of what's possible in logical reasoning! We’re looking for a motivated individual to build, maintain, and improve innovative approaches including (but not limited to!) energy-based modeling (EBM). You'll be at the forefront of designing and refining algorithms that go beyond the capabilities of traditional Large Language Models (LLMs), tackling complex reasoning challenges. You'll work closely with a talented team of AI experts, EBM specialists, formal verification engineers, and software developers to create groundbreaking solutions.
What you'll do
Research new reasoning algorithms and models
Pre-train and fine-tune the State-of-the-Art LLMs
Combine Reasoning algorithm and LLMs
Build effective and efficient ML pipelines
Collaborate with other teams to understand their pain points and priorities to define milestones of the corresponding roadmaps
Derive practical solutions and integrate them with the results of other teams to provide the best overall resolution
Qualifications
M.Sc. or Ph.D. (preferable) focusing on one or more of the following areas: Computer Science, Artificial Intelligence, Mathematics, or a closely related field
5+ years of subject matter expertise and research in one or more of the following areas: Machine Learning, Deep Learning, Reasoning, Energy-based Modeling (preferable)
Expertise in programming languages and tools critical for high-performance computing in Python/C++ and machine learning including Deep Learning frameworks like PyTorch /TensorFlow/JAX
Strong communication and teamwork skills
Experience with explicit and implicit reasonin
Demonstrated research publications in any of the major conferences (CVPR, ICLR, ICML, NeurIPS, ICCV, AAAI, ACL, etc.)
Bonus Points
Multi-node and multi-GPU training
Mathematical Reasoning – discrete math and logic
Provable record of Energy-based usage for different problems
Hands-on with algorithms used to train Energy-based models