What happened when they installed ChatGPT on a nuclear supercomputer
TL;DR
Los Alamos National Laboratory partnered with OpenAI to install ChatGPT on supercomputers used to process nuclear weapons testing data.
Key Points
- The collaboration is part of a broader program called 'Gemini' aimed at accelerating scientific research at the lab.
- The relationship between US nuclear weapons research and cutting-edge computing dates back to 1943, when physicists like Feynman ran human-vs-machine contests.
- AI tools are already reshaping how scientists at Los Alamos conduct research, from data analysis to simulation workflows.
Nauti's Take
Combining nuclear weapons infrastructure with a commercial LLM sounds like a screenplay Hollywood would have rejected – yet here we are. The move represents enormous institutional trust in OpenAI, but also a meaningful risk signal: LLMs hallucinate, lack full auditability, and were never designed for safety-critical nuclear applications.
The historical parallel to Feynman is romantic but misleading – back then, researchers knew exactly what the machine was doing. With GPT-4, that remains an open question.
Context
Los Alamos is no ordinary research lab – it is where the safety and reliability of the US nuclear arsenal is calculated and simulated. Granting a commercial large language model like ChatGPT access to this infrastructure raises serious questions about data security, auditability, and oversight. It also signals how deeply AI is now penetrating security-critical domains that were previously tightly insulated from commercial technology.
Tweets
Sources