Home / News / Technology / OpenAI Introduces Tailored Japanese GPT-4 Model as It Sets up Shop in Tokyo
Technology
3 min read

OpenAI Introduces Tailored Japanese GPT-4 Model as It Sets up Shop in Tokyo

Published April 15, 2024 4:15 PM
James Morales
Published April 15, 2024 4:15 PM

Key Takeaways

  • OpenAI has opened its 3rd international office in Tokyo.
  • To coincide with the opening, OpenAI has announced a new GPT-4 model customized for Japanese.
  • Japan’s AI scene is flourishing, attracting significant AI investment.

OpenAI has officially opened its first Asian office in Tokyo, Japan, marking a significant international expansion for the company, which also has offices in San Francisco and London.

To mark the occasion, the AI developer announced the release of a new custom GPT-4 model optimized for Japanese that could help it secure a role in the country’s burgeoning AI market.

English Bias in AI

While ChatGPT is already available in Japanese, the new model goes beyond OpenAI’s existing Japanese language services, which essentially run the standard GPT-4 model with an additional translation layer.

Although that approach may be sufficient for Spanish or French, which are structurally very similar to English, building a high-performance large language model for Japanese requires a greater degree of customization.

Part of the challenge AI developers face is the strong English bias in AI training data.

With over half of all websites using English as their main language and more English language books available in digital format than any other language, it was perhaps inevitable that language models would master English first.

Nevertheless, international AI developers have increasingly compiled libraries of non-English texts to help build a new generation of LLMs oriented toward other languages.

Improving Outcomes For Japanese AI Users

OpenAI claims  its new model can generate outputs in Japanese 3x faster than GPT-4 Turbo. And while it hasn’t been made generally available yet, users who were granted early access have already reported an improvement. 

According to Sunday’s blog post, after integrating the new custom GPT, the English learning app Speak can output tutor explanations 2.8x faster than before, with a 47% reduction in token cost.

The new model will certainly contribute to Japan’s emerging AI sector, but even before OpenAI’s arrival, the country was attracting significant AI investment.

Japan Draws AI Investment

Earlier this month, Microsoft announced its intention to invest $2.9 billion into Japan’s AI and cloud infrastructure over the next two years. Meanwhile, by 2027, Amazon Web Services will have invested nearly $18 billion in its Japanese data center operations.

Part of the country’s draw is its established technology sector and fertile AI research scene. But Japan’s regulatory environment also holds special appeal to firms like OpenAI.

With AI developers currently clashing with copyright owners in Europe and the US, Japan is one of the few countries that has explicitly amended  copyright law to clarify that AI training counts as fair use of intellectual property.

Was this Article helpful? Yes No