In the fiercely competitive field of artificial intelligence, U.S. companies like OpenAI and Google are locking horns to lead the market with large language models (LLMs). However, Japanese companies are not far behind in the race, spurred by the concept of economic security.
The global supply chain’s complexity has increased for producing goods and services, making it vulnerable to geopolitical crises. Japan’s foray into developing its own LLM is seen as a crucial step towards securing its economic infrastructure in the field of generative AI. With recommendations from the Liberal Democratic Party in May 2023, and a policy proposal from the Japan Business Federation (Keidanren) in June the same year, the emphasis on domestic development of advanced AI is evident.
Yet, the success of a national AI model is not solely predicated on security claims. Users often prioritize performance and cost over the origin of an LLM. No matter how committed the government is to promoting a ‘Japan-made LLM’, it will not succeed without meeting these key user demands.
Despite being a later entrant, Japanese companies are innovating their development techniques to claim their stake in the AI model development competition. This includes start-up ELYZA, whose CEO Yuya Soneoka claims to have achieved a globally competitive level with their new language model.
ELYZA’s latest model boasts of superior language abilities in Japanese, enhanced by their unique data on top of Meta’s open-source LLM ‘Llama 2’. With 70 billion parameters, their model performed impressively against global counterparts in tests, ranking fourth in writing and inferencing tasks.
Nevertheless, Yuya Soneoka acknowledges the gap between local and international models, with Japan’s best still only just debuting at a commercial standard. To remain competitive against the likes of Google and Microsoft, essential resources and distribution networks are critical.
In a strategic move, ELYZA has embraced a subsidiary relationship with KDDI. This partnership aims to overcome the barriers to business adoption, such as operational computing resources and sales networks. Soneoka candidly addressed the funding challenges that could arise without such collaboration, emphasizing the importance of joining forces with established entities.
Importance of Economic Security in AI Development
Economic security is becoming increasingly important as nations vie for technological independence and resilience in the AI domain. Japan’s drive to create its own AI models stems from a desire to reduce dependence on foreign technology, which is a concentration of power in the AI sector currently dominated by American and Chinese entities.
User Demands vs. National Goals
While Japan aspires to promote domestically produced LLMs, users are generally more interested in the cost, efficiency, and functionality of these technologies. A ‘Japan-made’ label alone will not suffice if the performance does not match or surpass that of competitors.
Innovation Through Partnership
Japanese startups like ELYZA are leveraging strategic partnerships with larger corporations like KDDI to address the challenges of resource allocation and market penetration. Relationships with established firms offer start-ups valuable infrastructure, funding, and distribution networks, which are integral to achieving competitive standing in the global market.
Key Challenges and Controversies
A prevalent challenge in AI development is the ability to sustain an innovative edge while managing costs, something large entities like Google and Microsoft excel at due to their financial and computational resources. These challenges are amplified for smaller economies and companies. Additionally, AI ethics and privacy are hot topics, but specific controversies around Japan’s AI models are not mentioned in the article.
Advantages and Disadvantages
Advantages of Japan developing competitive AI language models include technological sovereignty, customized solutions tailored to the Japanese language and culture, and the growth of local tech ecosystems. However, disadvantages encompass the high cost of R&D, late entry to an established market, and potential lack of global collaborative opportunities that could enhance learning and development.
For more information on the topic, you may visit the sites of Japanese businesses developing AI technologies or the official websites of their partner corporations. As requested, I am providing links to the main domains only:
– Japan Business Federation (Keidanren)
– KDDI Corporation
These URLs are provided with the assumption of their validity based on their relevance to the mentioned entities in the article.
The source of the article is from the blog anexartiti.gr