High-tech companies are betting that nuclear power will help deliver the electricity it needs to make its AI plans come true. However, data centers need electricity tomorrow, and the nuclear industry is not known for its speed.
Trey Lauderdale believes that AI can give the nucleus the speed it needs.
Lauderdale’s obsession with nuclear power began near the home. In San Luis Obispo, California, where he lives, he continued to run to those who worked at the Diablo Canyon Power Plant. “They’re like coaches on our flag football team,” he said.
Speaking to them, he learned that the nuclear power plant was swimming in writing. Diablo Canyon is worth around 2 billion pages, he said. Serial healthcare entrepreneur Lauderdale had a feeling that AI could help the nuclear industry maintain paper problems.
Lauderdale founded Atomic Canyon a year and a half ago, and initially funded it with his own money. Startups use AI to help engineers, maintenance technicians and compliance personnel find the documents they need.
The startup signed a contract with Diablo Canyon in late 2024. Lauderdale said the agreement led to inquiries from other nuclear companies. “That was when we were at the point where we needed to raise capital as entrepreneurs.”
Atomic Canyon has closed its $7 million seed round led by Energy Impact Partners, the company told TechCrunch exclusively. Participating investors include Commonweal Ventures, Plag and Play Ventures, Tower Research Ventures, Wischoff Ventures, and former angel investors.
When Atomic Canyon first began, its AI engineers tested a variety of models with overwhelming results. “We quickly realized that when we saw these nuclear words, AI hallucinates,” Lauderdale said. “We haven’t seen enough examples of acronyms.”
However, building new AI models requires large computing power. So, Lauderdale conducted nuclear research and spoke about his way to a meeting with Oak Ridge National Laboratory, which has the world’s second fastest supercomputer. The lab was intrigued by the idea and was awarded an Atomic Canyon 20,000 GPU time computing.
The Atomic Canyon model uses statement embeddings that are particularly suitable for indexing documents. It leaves nuclear power plant documents to the creation of searchable documents using searched high-end generation or rags. Rag uses a large language model to create responses to queries, while LLMS must refer to specific documents to reduce hallucinations.
For now, Atomic Canyon is sticking to documenting searches due to its low stakes.
“One of the reasons we start a generative work centered around document titles is that it can lead to someone a bit of frustration by making it wrong.
Ultimately, Lauderdale expects to create a “first round draft” of “Atomic Canyon AI” with References and a “first round draft” of the document. “You’re always going to have a human in the loop here,” he said.
However, Lauderdale did not place a timeline on that effort. Searches are “basic layers,” he said. “You need to nail the search,” he said, furthermore, given the number of documents in the nuclear industry, “there are long runways during the search.”
Source link
