Official announcements from major scientific award committees

AMD and Johns Hopkins Unveil Agent Laboratory with 84% Cost Reduction, 95.7% Success Rate

Jan 9, 2025, 06:51 AM

AMD researchers, in collaboration with Johns Hopkins, have introduced 'Agent Laboratory,' an autonomous framework leveraging large language models (LLMs) to streamline the scientific research process. The system automates key stages such as literature review, experimentation, and report writing, enabling researchers to focus on ideation and critical thinking. The framework features specialized agents, including 'PhD agents' for literature reviews and 'ML Engineer agents' for machine learning code generation. Initial tests, led by Schmidgall et al., have shown that the system, particularly when using the o1-preview backend, produces high-quality research outputs, including state-of-the-art machine learning code. Additionally, it significantly reduces research costs by up to 84%, with some projects completed for as low as $2.33, and achieves a 95.7% success rate across tasks. The tool is open-source, adaptable to various computing power levels, and designed to complement researchers rather than replace them. Human involvement at each stage improves the overall quality of research.