Fujitsu invested in Cohere as part of a partnership to develop an LLM for the underserved language.
Japanese IT services company Fujitsu has launched Takane, a Japanese-language large language model (LLM) powered by Toronto-based enterprise artificial intelligence (AI) startup Cohere.
Fujitsu claims Takane has achieved industry-leading performance in multiple Japanese-language benchmarks.
The new LLM is a product of a partnership struck between the two companies earlier this year, in which the companies agreed to develop a Japanese LLM for private cloud usage. As a result, Fujitsu made “a significant investment” into Cohere’s $500 million USD ($687 million CAD) Series D round in July.
According to Fujitsu, Takane is designed for enterprise use in secure private environments, such as the finance, manufacturing, and the security sectors, and will be integrated into its Kozuchi, data intelligence, and Uvance platforms.
“We are very excited to bring Takane’s advanced Japanese LLMs to global enterprises,” Cohere CEO Aidan Gomez said in a statement. “Our partnership with Fujitsu accelerates AI adoption in this critically important market by offering secure, performant AI designed specifically for business use across Japanese and other languages.”
Cohere released Command R+, the LLM Takane is based on, earlier this year. The model had a focus on multilingual use, which Gomez said is “crucial for equitable utility of this technology” at the time. Studies have shown that LLMs are largely trained on English language data and, as a result, are less likely to understand other languages like Japanese.
RELATED: Cohere and Float once again top LinkedIn’s Top Canadian Startups list
Fujitsu claims that, with additional training and fine-tuning, Takane has achieved industry-leading performance in multiple Japanese-language benchmarks language understanding, natural language inference, reading comprehension tasks, semantic understanding, and syntactic analysis.
In a statement, Fujitsu said that characteristics of the Japanese language, including the mixed use of multiple character types, omitted subjects, and honorific expressions, have posed significant hurdles for general-purpose LLMs. The company added that an accurate Japanese LLM is important for the government, finance, healthcare, and law sectors where “minor language errors can have serious consequences.”
Cohere has become one of Canada’s largest AI players, having topped LinkedIn’s annual list of top Canadian startups for the second time last week. The list is based on the platform’s data on employee growth, jobseeker interest, engagement within the company and its employees, and talent attraction.
Feature image courtesy Cohere.
