**Japan’s Push to Develop Japanese-Language AI Chatbots**
Japan is investing heavily in the development of its own versions of ChatGPT, the popular AI chatbot created by OpenAI. The Japanese government, along with technology giants NEC, Fujitsu, and SoftBank, is dedicating significant resources to building AI systems based on large language models (LLMs) but trained specifically for the Japanese language. While current LLMs, such as GPT-3, excel in English, they often struggle to accurately process Japanese due to differences in the alphabet system, limited data, and other factors.
LLMs rely on massive amounts of publicly available data to learn the patterns of natural speech and prose. However, since GPT-3’s training data mostly consists of English text, using it to process Japanese queries requires translating the Japanese text into English, finding the answer, and then translating it back into Japanese. This process often leads to incorrect translations, as Japanese sentence structure differs significantly from English.
Moreover, Japanese LLMs must also consider cultural practices and language nuances to accurately respond to queries. For instance, if ChatGPT were to generate a job-application email in Japanese, it might omit standard expressions of politeness, making it appear as an obvious translation from English. Researchers have developed a ranking system, called Rakuda, to assess how well LLMs can answer open-ended questions about Japanese topics. While Japanese LLMs are improving, they still trail behind GPT-4.
To address this issue, Japan is undertaking several initiatives to develop robust Japanese LLMs. One significant effort involves using the Japanese supercomputer Fugaku, one of the world’s fastest, to train an LLM on Japanese-language input. This project, led by the Tokyo Institute of Technology, Tohoku University, Fujitsu, and the government-funded RIKEN research centers, aims to release the LLM next year as an open-source model. The team hopes to give the LLM at least 30 billion parameters, making it comparable in size to other LLMs.
Additionally, the Japanese Ministry of Education, Culture, Sports, Science, and Technology is funding the creation of a Japanese AI program focused on scientific research. This program will generate scientific hypotheses by learning from published research, speeding up target identification for further investigation. The LLM’s initial size is projected to be 100 billion parameters and could cost at least ¥30 billion ($204 million) to develop.
Other Japanese companies, such as NEC and SoftBank, are also venturing into the commercialization of their own Japanese-language LLM technologies. NEC, for example, has reduced the time required to create internal reports and software source code using its generative AI. SoftBank plans to invest ¥20 billion into generative AI trained on Japanese text and will launch its LLM next year, focusing on helping companies digitize their operations and increase productivity.
Japanese researchers anticipate that the development of accurate and effective Japanese-language AI chatbots will greatly impact language learning and research on Japan. By bridging the gap between Japan and the rest of the world, these chatbots have the potential to advance international joint research.
**Editor Notes: Encouraging Japanese-Language AI Development**
The development of Japanese-language AI chatbots marks a significant step for Japan in bridging the gap between its language and culture and AI capabilities. The unique complexities of the Japanese language present challenges that traditional English-trained language models often struggle to overcome. However, Japan’s investments in building its own LLMs tailored to the Japanese language are promising.
By combining the power of large language models with the intricacies of the Japanese language and culture, these Japanese AI chatbots have the potential to enhance language learning, improve research on Japan, and foster international collaborations. As more resources are dedicated to these developments, we can expect to see even more accurate and effective AI systems that cater to the specific needs of the Japanese language.
To stay updated on the latest advancements in AI and the impact of language models, visit [GPT News Room](https://gptnewsroom.com), where you can find comprehensive news and insights in the field.
Source link
from GPT News Room https://ift.tt/o2RXain
No comments:
Post a Comment