Los Alamos National Laboratory teams up with OpenAI to improve AI safety
Researchers at Los Alamos National Laboratory are collaborating with OpenAI on an evaluation study to enhance artificial intelligence safety. This pioneering evaluation aims to contribute to cutting-edge research on AI biosecurity assessments.
Advancing AI Capabilities Responsibly
"The potential benefits of advancing AI capabilities are vast," stated Erick LeBrun, a research scientist at Los Alamos. He emphasized the importance of understanding and measuring any potential risks or misuses of advanced AI technology in relation to biological threats. The collaboration with OpenAI represents a significant step towards establishing a framework for assessing current and future AI models, ensuring the responsible development and deployment of AI technologies.
AI-enabled biological threats could present a substantial risk, and the team is focused on exploring how multimodal, frontier models could potentially lower the barrier for non-experts to create such threats. The upcoming evaluation builds upon previous work and aligns with OpenAI’s Preparedness Framework, which guides the tracking, evaluation, forecasting, and protection against emerging biological risks.
Evaluating ChatGPT-4 for Biological Tasks
In earlier evaluations, the research team observed that ChatGPT-4 showed a slight improvement in providing information relevant to the creation of biological threats. However, these experiments primarily focused on human performance in written tasks rather than biological benchwork. Furthermore, the model inputs and outputs were text-based, excluding vision and voice data.
The upcoming evaluation will be the first experiment to assess multimodal frontier models in a laboratory setting, using proxy tasks and materials. Experts' abilities to execute and troubleshoot a safe protocol involving standard laboratory experimental tasks will be evaluated. The team aims to quantify the impact of ChatGPT-4 on task completion and accuracy, thereby assessing how frontier models can assist in real-world biological tasks.
Partnership for Advancing Scientific Research
Mira Murati, OpenAI’s Chief Technology Officer, expressed excitement about the partnership with Los Alamos National Laboratory, emphasizing the significance of this collaboration in advancing scientific research and addressing potential risks associated with AI technology. This partnership aligns with the recent White House Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.
For more information and feature stories from Los Alamos National Laboratory, you can sign up for updates on their website.
Los Alamos National Laboratory
P.O. Box 1663 Los Alamos, NM 87545
(505) 667-5061
Managed by Triad National Security, LLC for the U.S. Dept. of Energy’s NNSA
Copyright 2024 Triad National Security, LLC. All Rights Reserved.
Learn about the Department of Energy’s Vulnerability Disclosure Program
