Google Unveils LaMDA and MUM AI Models for Natural Dialogue
Google introduces LaMDA, an open-domain language model built for natural conversation, alongside MUM, a multimodal AI that understands text and images across 75 languages to answer complex questions.
Google unveils LaMDA, a next-generation language model designed specifically for dialogue applications. Unlike current leading models, LaMDA excels at understanding conversation context and does not require retraining to discuss new topics. The open-domain model already possesses vast knowledge about subjects like Pluto and millions of other areas without needing predefined responses or hand-programmed concepts.
LaMDA generates natural, open-ended conversations that never follow the exact same path twice. However, Google CEO Sundar Pichai admits the technology remains in the research and development phase, meaning it occasionally produces nonsensical answers. Despite these early imperfections, the model shows impressive potential for carrying on fluid conversations about virtually any subject.
In a related announcement, Google details MUM, a multimodal model trained across 75 languages that simultaneously understands text, images, and videos. MUM tackles complex queries by connecting contextual dots, such as realizing that preparing for a Mount Fuji hike involves both fitness training and checking weather conditions. The model even analyzes images to answer practical questions, like determining whether a specific pair of hiking boots is suitable for the journey.