The NIKKEI newspaper announced on the 24th that it has developed foundational technology for an AI (artificial intelligence) specializing in economic information. The technology, known as a large-scale language model, has been trained on approximately 40 years’ worth of articles from the NIKKEI Group’s newspapers and magazines. It is anticipated to be utilized for functions such as article summarization.
The large-scale language model is called the “NIKKEI Language Model,” and there are two types with parameters numbers of 130 billion and 700 billion, respectively. It was developed by NIKKEI’s research and development organization, NIKKEI Innovation Lab.
The model with a maximum of 130 billion parameters is not based on existing models generally available to the public but has been independently constructed by NIKKEI. It was trained only on data owned by the NIKKEI Group, including articles from the NIKKEI newspaper, electronic versions, specialized papers like NIKKEI MJ, and various media issued by NIKKEI BP.
https://www.nikkei.com/article/DGXZQOUC1941R0Z10C24A4000000
Editor’s view
NIKKEI is a well-established business magazine in Japan and has recently been steadily reforming itself as a DX and technology company.
NIKKEI has a vast amount of data in the area of Japan x economy, and we have high expectations for this new generation AI as a specialized type of AI, unlike general-purpose generation AI.
However, Nikkei has publicly announced its policy that “responsible reporting is handled by people,” and does not use AI to create or edit newspaper articles.
I would like to commend the company’s efforts and courage in developing a generative AI in the face of such a paradox.
