Japan's IT giant Fujitsu has recently unveiled its latest AI strategy, focusing not on developing Generative AI (GenAI) to compete with other companies, but on creating a platform that enables businesses to easily customize their own GenAI solutions.
Fujitsu stated that its newly developed GenAI Framework for Enterprises is designed to address challenges facing enterprises when applying GenAI, such as the difficulty in acquiring large amounts of training data, the inability to meet diverse cost and processing speed requirements, and the potential for non-compliance with laws and regulations. This framework will be available to global customers starting in July 2024 through the company's cloud-based AI platform, Fujitsu Kozuchi.
In response to the issues identified by enterprises in testing GenAI in 2023, Fujitsu has incorporated three key technologies into the GenAI Framework for Enterprises: Knowledge Graph Extended Retrieval-Augmented Generation (RAG), GenAI Amalgamation, and GenAI Auditing. RAG is a commonly used technique for customizing GenAI models for enterprise applications.
However, existing RAG approaches suffer from fragmented and disconnected knowledge learning, which can lead to errors in real-world applications. For instance, a RAG model might reference different and incompatible repair manuals for different car models, or misquote legal provisions, confusing the meanings of prohibited, not recommended, and encouraged actions.
Fujitsu's Knowledge Graph Extended RAG addresses these limitations by using knowledge graphs to contextualize learning data. The model can handle over 10 million tokens, the basic units of text data, allowing it to learn from large amounts of complex data and generate highly accurate outputs.
According to Fujitsu, Knowledge Graph Extended RAG can reduce the amount of training data required by up to 25%. Moreover, this model has been recognized for its world-leading accuracy by the Association for Computational Linguistics (ACL), a leading international conference in the field of natural language processing.
Beyond improvements in RAG, Fujitsu also offers a platform for various GenAI hybrid applications. This allows users to easily adopt suitable GenAI models, reducing the working hours required for GenAI customization by 30% and increasing operational efficiency by 25%.
Fujitsu's Knowledge Graph Extended RAG, which improves learning efficiency, also provides real-time monitoring of the results of multiple GenAI hybrid applications. It offers logical reasoning support, enabling users to easily determine whether the results are applicable.
