Here's why Japan is gearing up to build its own ChatGPT

“Current public LLMs, such as GPT, excel in English, but often fall short in Japanese due to differences in the alphabet system, limited data and other factors,” Keisuke Sakaguchi told Nature. He is a researcher at Tohoku University in Japan who specializes in natural language processing.

Why Japan needs its own ChatGPT

Large language models (LLMs) are trained on huge amounts of data – zettabytes and yottabytes of data. The LLM that ChatGPT is based on – GPT-4 – has been trained mostly in the English language. Scientists are concerned that these LLMs cannot grasp the intricacies of Japanese culture and language. 

English has 26 letters, while written Japanese consists of two sets of 48 basic characters plus 2,136 regularly used Chinese characters or kanji. Most kanji have two or more pronunciations and a further 50,000 or so rarely used kanji exist. This could be a complex issue for ChatGPT. Hence, the need for Japan’s own ChatGPT-like chatbot.

Big tech firms like NEC, Fujitsu, and SoftBank are spending millions of dollars to create artificial intelligence tools; however, the country lacks technological knowledge, making it fall behind countries like the US, China, and the EU, all of which are at the forefront of developing algorithms.

It is estimated that by 2030, Japan will face a deficit of 789,000 software engineers, according to the Ministry of Economy Trade and Industry, reported CNBC. The reason behind the need for more engineers is the same as in every other industry in Japan: there are fewer people who can work due to an aging population.

Lack of technical know-how and manpower

As per IMD World Digital Competitiveness Ranking, Japan is ranked 28 out of 63 countries in terms of technological knowledge. The country also has a shortage of formidable AI supercomputers, which are needed to train an AI tool with vast amounts of data. But no private Japanese company possesses a “world-class machine” with that kind of capabilities, reported Nikkei Asia. 

But while Japan may not possess the technology or the manpower to build the latest AI tech, it is only a matter of resources. “Certainly Japanese LLMs are getting much better, but they are far behind GPT-4,” said Sam Passaglia, a physicist at the University of Tokyo who studies Japanese language models. Passaglia and his colleagues launched Rakuda, which is used for ranking and benchmarking Japanese AI assistants.

Government supercomputers like Fugaku “hold the key” to Japan’s pursuit of LLMs, said Noriyuki Kojima, co-founder of Japanese LLM startup Kotoba Technology, in an interview with CNBC. “Access to such large-scale supercomputers forms the backbone of LLM development, as it has traditionally been the most significant bottleneck in the process,” he added.