SoftBank launches an OpenAI for Japan: SB Intuitions, building LLMs and generative AI in Japanese


AI may have found a center of gravity in the Bay Area, but we’re seeing a lot of activity elsewhere among those who believe that there is a market opportunity ensuring that the pace of innovation does not only get played out in just one corner of the world.

Today, SoftBank put in a bid for businesses and consumers in Japan with the launch of SB Intuitions, a new company that it says will “research and develop homegrown Large Language Models (LLM) specialized for the Japanese language.” SB Intuitions also plans to build and sell generative AI services based on Japanese.

The announcement was short on detail but for now, here’s what we know. SB Intuitions will be 100% owned by SoftBank itself — no call for or plans at this point for outside investors. SoftBank will only use data housed in Japan-based data centers — but no detail so far on how it will source the data and under what agreements. It is likely to tap into the company’s extensive consumer and enterprise operations in the country.

The earliest plans for this business were actually laid out in March of this year; but the SB Intuitions name and the company’s plans are only getting announced today. To operate the service, SoftBank will be tapping into another big AI effort at the company: a computing platform, built on Nvidia GPUs, that “can be utilized for developing generative AI and other applications,” including academic research and enterprise services. That will be online before the end of March 2024, the company said.

Hironobu Tamba — a long-time SoftBank employee who has most recently been running the company’s AI and IoT division — will lead the new business, which is starting off with a paid-in capital sum of 150 million yen — or just over $1 million at today’s rates. That appears to be a basic sum to establish the business; SoftBank is not disclosing how much it is going to invest in the effort overall.

But considering that OpenAI has raised over $11 billion and is valued at nearly $30 billion as of its last raise this year, that gives you an idea of the sums of money needed by the most ambitious companies building LLMs and generative AI services.

SoftBank separately in its earnings presentation said that it would be issuing a new bond for 120 billion yen (just over $840 million) with some of the proceeds going to AI investments, so that could point to what it will be tapping for SB Intuitions in the future.

It’s interesting and a little ironic to see SoftBank launching a new LLM business right now, given the hit-and-miss relationship SoftBank Group has had with AI over the years.

SoftBank the tech company presciently identified AI as an important area for development years ago, well before the current wave of AI hype started to take off. But it’s hard to find any evidence or success of the services it built in-house and the partnerships it forged to get that strategy off the ground. (Here is one example of what I mean. Whatever happened to this?)

The same might be said of its activity as an AI investor. Overall, SoftBank has been very, very bullish on investing in AI, with this article in the WSJ estimating that it had spent some $140 billion on some 400 investments over the years to play a role there. Its biggest effort as an investor by a wide margin has been the Vision Fund, which over multiple funds has raised more than $150 billion to invest in startups, with AI a huge focus. The WSJ’s conclusion from that, however, was damning: Despite all that money spent, SoftBank is virtually cut out of the current wave of AI developments.

SoftBank’s efforts are targeted partly at that lost business opportunity but also a bigger, existential concept.

Just as search engines, operating systems and the cloud have all been major milestones in how the world has engaged with the internet, “we believe that generative AI will be a technology of incomparable importance,” SoftBank CEO and president Junichi Miyakawa said in the presentation today.

“However, if you look at the current situation, the only major players developing large language models are foreign companies. These generative AIs are being developed based on English and Chinese datasets and information, and thought patterns based on these languages have been built up,” he added. “The significance of… a domestically-produced a generative AI is that it will be developed on Japanese language data set making it more suited to Japanese business practice and culture. In particular, we believe that it will be able to handle unique expressions in Japanese, such as those used in public and medical services.”

This work will not be carried out in a bubble. In fact, in the presentation today, Tamba also highlighted that SoftBank has a strategic alliance with Microsoft that will involve SoftBank providing a secure data environment for enterprises in Japan that might be interested in working with Microsoft on AI, and generative AI, initiatives. SoftBank itself will also be using Microsoft’s Copilot service internally.

“We will not only develop our own large language models, but also establish a multi-generative AI system in which we select the most appropriate model for our clients’ needs from among multiple models developed by OpenAI, Microsoft, Google and other companies,” said the CEO.



Source link