OpenAI and Los Alamos National Laboratory Collaborate to Strengthen AI Safety
OpenAI, an AI company known for developing AI models such as
OpenAI and Los Alamos National Laboratory announce research partnership | OpenAI
https://openai.com/index/openai-and-los-alamos-national-laboratory-work-together/
Los Alamos National Laboratory teams up with OpenAI to improve frontier model safety
'As a private organization committed to serving the public good, we're excited to announce our first partnership with Los Alamos National Laboratory to explore the potential of biosciences,' said OpenAI CTO Mira Murati about the partnership. 'This partnership represents a natural extension of our mission to advance scientific research while understanding and reducing risk.'
'AI is a powerful tool that has the potential to deliver great scientific benefits, but like any emerging technology, it comes with risks,' said Nick Generas, deputy leader of the Information Systems and Modeling Division at Los Alamos National Laboratory. 'This work will be led by our new AI Risks and Technology Assessment Group, which will help assess and better understand these risks.'
According to Los Alamos National Laboratory, AI-enabled biological threats could pose a significant risk, but existing research has not assessed how multimodal future AI models (frontier models) would lower the hurdle for non-experts to create biological threats.
This effort establishes a framework for evaluating current and future AI models and is an important step toward ensuring the responsible development and deployment of AI technologies, explained Eric LeBlanc of Los Alamos National Laboratory.
Already, OpenAI's research team has found that GPT-4-based ChatGPT can provide information that could generate biological threats, but the experiment was limited to text as input and output for the model, and did not include images or audio.
In the future, they plan to quantify and evaluate how frontier models assist biological tasks by examining the improvement in accuracy of tasks performed by GPT-4o-based ChatGPT.
Related Posts:
in Software, Posted by logc_nt