Google AI Research and DeepMind have released VaultGemma 1B, the largest open-weight large language model trained entirely with differential privacy (DP). This development is a major step toward building AI models that are both powerful and privacy-preserving. Large language models trained on vast web-scale datasets are prone to memorizing sensitive information from the training data, raising concerns about user privacy and data security. Differential privacy provides a framework for quantifying the privacy guarantees provided by an algorithm or system. By training models with differential privacy, sensitive information in the training data is protected, enabling the development of powerful AI models while preserving user privacy. VaultGemma 1B is significant not only in its scale but also in its focus on privacy, setting a new standard for large language models. The release of VaultGemma 1B represents a milestone in the development of privacy-preserving AI technologies and highlights Google AI’s commitment to advancing the field in a responsible and ethical manner.

<Mark Tech Post 뉴스 본문 전체읽기>