TAU Intelligence is excited to announce a promising partnership with Jardine Matheson and Intel. This strategic collaboration aims to advance the application of Generative AI by introducing two pioneering proofs-of-concept (PoCs) for Jardines, focusing on legal contract review and customer service.
TAU Intelligence employs a moderately sized large language model (LLM), which the Jardines team customizes with on-premises fine-tuning. This distinct approach differentiates this collaboration from the common methods used in many other AI projects. Crucially, it ensures that data is neither used for training nor retained by the LLM provider.
Initial findings indicate that this approach maintains exceptional accuracy and output quality when reviewing and analyzing legal contract provisions and common user queries. Moreover, it demonstrates notable efficiency in terms of energy consumption and computing resources.
Since 2022, TAU’s extensive research and development efforts have underscored the challenges posed by large LLMs, particularly their high costs, data security concerns, and environmental impact. Other studies indicate that popular LLMs consume 15 times more energy than a typical web search, raising environmental concerns. TAU’s strategy of using smaller and domain-specific LLMs that operate locally within a company’s own environment addresses these issues, ensuring resource-efficient AI and data sovereignty.
The solutions created through this partnership will undergo testing on Intel’s upcoming Xeon processors designed for data centers, cloud computing, and edge computing tasks. Leveraging these processors, the collaboration aims to achieve substantial performance enhancements, utilizing Intel's leading accelerator engines and improving performance per watt across critical workloads such as AI, data analytics, high-performance computing (HPC), and more.
Professor Lam Kwok Yan, Co-founder of TAU Intelligence and a renowned AI Safety researcher based in Singapore, commented, "The approach on these PoCs is distinct from the current trend where LLMs are rapidly expanding in size and consumption of energy and computing power. By focusing on domain-specific knowledge, we can fine-tune AI models in order to deploy smaller LLMs in a safe manner, mitigating the problem of hallucination and ensuring the responsible use of AI. The collaboration with Jardines signifies a focus on the development of responsible and resource-efficient AI solutions by companies, setting a benchmark for innovation and sustainability in the AI industry”.
Anne O’Riordan, Group Digital Director at Jardine Matheson, emphasized the significance of this partnership, “This collaboration highlights Jardines’ ongoing efforts to develop, test, and deploy AI to position our businesses for the future. These PoCs tackle real business challenges, enhancing operations by combining high-quality output with resource efficiency – while ensuring the security of our proprietary data. We’re excited to see the results of this engagement and by the potential for future collaboration”.