Atomic Canyon: AI for the Nuclear Industry - ChatGPT Alternative

AI and the Future of Nuclear Power

Technology firms are making substantial investments, anticipating that nuclear energy will be crucial in supplying the electricity needed for their artificial intelligence initiatives. However, data centers require immediate power access, a demand that clashes with the traditionally slow pace of development within the nuclear sector.

Addressing the Speed Challenge

Trey Lauderdale believes that advancements in AI can provide the nuclear industry with the accelerated capabilities it requires. His interest in nuclear technology originated from his local community.

Living in San Luis Obispo, California, Lauderdale frequently encountered individuals employed at the Diablo Canyon Power Plant. He described them as integral figures within the local community, similar to sports coaches.

The Document Challenge

Through conversations with these professionals, Lauderdale discovered the immense volume of documentation within nuclear power plants. He estimates that Diablo Canyon alone possesses approximately 2 billion pages of documentation.

As a seasoned healthcare entrepreneur, Lauderdale hypothesized that AI could effectively manage this extensive paperwork challenge faced by the nuclear industry.

Atomic Canyon's Founding and Initial Funding

Lauderdale established Atomic Canyon just over a year and a half ago, initially self-funding the venture. The company focuses on utilizing AI to assist engineers, maintenance personnel, and compliance officers in locating necessary documents.

A contract secured with Diablo Canyon in late 2024 spurred interest from other nuclear power companies. Lauderdale recognized this as a pivotal moment, signaling the need for external capital.

Seed Funding Round

Atomic Canyon successfully closed a $7 million seed funding round, led by Energy Impact Partners. Additional investors included Commonweal Ventures, Plug and Play Ventures, Tower Research Ventures, Wischoff Ventures, and existing angel investors.

Overcoming AI Hallucinations

Early testing of AI models by Atomic Canyon yielded disappointing results. The AI exhibited a tendency to “hallucinate” when processing specialized nuclear terminology. This was attributed to a lack of sufficient training data containing relevant acronyms and industry-specific language.

Collaboration with Oak Ridge National Laboratory

Recognizing the need for substantial computing power to build a specialized AI model, Lauderdale secured a meeting with Oak Ridge National Laboratory. The laboratory, renowned for its nuclear research and possessing the world’s second-fastest supercomputer, was intrigued by the proposal.

Consequently, Atomic Canyon was granted access to 20,000 GPU hours of computing resources.

Utilizing Sentence Embedding and RAG

Atomic Canyon’s models employ sentence embedding, a technique well-suited for document indexing. The system utilizes retrieval-augmented generation (RAG) to make a nuclear power plant’s documents searchable.

RAG leverages large language models to formulate responses to inquiries, while simultaneously referencing specific documents to minimize the risk of inaccurate or fabricated information.

Focus on Document Search

Currently, Atomic Canyon is concentrating on enhancing document search capabilities, prioritizing a lower-risk approach.

Lauderdale explained that errors in generating document titles might cause minor inconvenience, but would not compromise safety at the plant.

Future Vision: AI-Assisted Document Creation

Looking ahead, Lauderdale envisions Atomic Canyon’s AI generating initial drafts of documents, complete with appropriate references. He emphasizes that human oversight will remain a critical component of the process.

While a specific timeline for this development remains undefined, Lauderdale asserts that perfecting search functionality is paramount. He also highlights the vast potential within document search alone, given the sheer volume of information in the nuclear industry.