TL;DR:
- NTT unveils its All Photonics Network (APN) technology as a core component of the Innovative Optical & Wireless Network (IOWN) concept.
- Shingo Kinoshita, Senior VP at NTT, explains the revolutionary APN technology, featuring an ordinary LSI chip with exceptional speed and low power consumption.
- NTT enters the realm of Large Language Model (LLM) technologies with its gen-AI model, “tsuzumi,” emphasizing collaboration with natural language and the potential for AI protocols.
- Challenges in LLM interactions, knowledge integration, and various protocols are highlighted, emphasizing the need for diverse approaches.
- NTT aims to expand beyond the Japanese language, including English, Chinese, Korean, and European languages.
- The lightweight tsuzumi LLM holds promise for applications in smartphones and car navigation systems.
- RakuDA benchmark is discussed, with tsuzumi lagging behind GPT3 in coding but determined to improve.
Main AI News:
NTT, a global leader in telecommunications and information technology services, recently unveiled its groundbreaking vision for the All Photonics Network (APN) technology during the NTT R&D Forum conference in Tokyo. This technological marvel is one of the pillars of the Innovative Optical & Wireless Network (IOWN) concept, alongside its cognitive foundation approach and digital twin computing.
Shingo Kinoshita, Senior VP & Head of Research and Development Planning for NTT, shed light on the intricate workings of these cutting-edge technologies. At the heart of the APN technology lies an ordinary yet extraordinary Large Scale Integration (LSI) chip. This chip possesses the remarkable capability to convert optical fiber and electrical wiring into light and electricity, achieving exceptional speeds while consuming minimal power.
In a stride that mirrors industry trends, NTT has ventured into the realm of Large Language Model (LLM) technologies for generative-AI applications. They’ve developed their own gen-AI model named “tsuzumi,” drawing inspiration from the traditional Japanese drum of the same name. The journey to create tsuzumi has been a formidable one.
Kinoshita elucidated the challenges of LLM collaboration, emphasizing the importance of the language’s natural integration with LLMs. However, he also hinted at the possibility of AI evolving its own protocol for enhanced efficiency.
Navigating the intricate web of LLM interactions, knowledge bases, and unique characteristics remains a formidable task. Kinoshita emphasized the need for diverse approaches, involving not only technology but also social science.
When questioned about NTT’s expansion plans, Kinoshita expressed enthusiasm for broadening their horizons beyond the Japanese language. Their global ambitions encompass English, Chinese, Korean, and various European languages.
Highlighting potential applications, Kinoshita emphasized the importance of lightweight and swift responsiveness, particularly for smartphones and car navigation systems. The nimble tsuzumi LLM could find its place in real-world deployments.
In the realm of benchmarks, Kinoshita acknowledged the challenges of relying on a single metric. He mentioned RakuDA, a widely used benchmark that encompasses a range of topics from history and geography to coding. While Tsuzumi currently lags behind GPT3 in coding proficiency, NTT remains committed to strengthening its capabilities, demonstrating its dedication to pushing the boundaries of LLM technology.
Conclusion:
NTT’s pioneering advances in photonics-based computing, along with their foray into LLM technologies, demonstrate their commitment to shaping the future of telecommunications and AI. Their global ambitions and dedication to overcoming LLM challenges position them as a key player in the evolving market, with the potential to revolutionize how we interact with technology and language models.
