Synopsis: In this AI Leadership Insights interview, Amanda Razani speaks with George Apostopoulos, founding engineer of Endor Labs, about the release of a new Endor AI tool that helps companies with choosing the best and safest open source AI models to use, as well as what key concerns business leaders have when it comes to selecting the right model.
In this interview, Amanda Razani speaks with George Apostopoulos, founding engineer at Endor Labs, about their innovative tool designed to help businesses evaluate and select AI models from platforms like Hugging Face. Apostopoulos explains how the tool, which provides “vendor scores” for AI models, simplifies the process for developers and company leaders to identify suitable models by considering factors such as licensing, popularity and security risks. By offering a natural language interface, the tool enables users to pose specific queries about model attributes, such as compatibility with open-source licensing or risk of vulnerabilities. Apostopoulos likens AI models to software dependencies, emphasizing the operational and security risks they may introduce, including legal issues, improper functionality, and potential vulnerabilities in model weight encoding.
Apostopoulos also addresses the broader decision-making process for businesses choosing between open-source and commercial AI models or creating fine-tuned models tailored to their own data. He highlights the advantages of locally fine-tuned open-source models for maintaining data privacy but notes the challenges in setting up infrastructure to manage them. He concludes with a warning about the underestimated risks associated with AI models, emphasizing the importance of staying vigilant against malicious actors as the technology continues to grow. The conversation sheds light on the complexities of integrating AI safely and effectively into modern organizations.
