Japan has not had a globally popular software app (outside of gaming) in decades. In the software realm, Japan is far, far behind still and I see no reasonably fast way for Japan to leapfrog like China did skipping the PC paradigm and focusing on mobile first.
Japan has kept a key spot in the semiconductor manufacturing supply chain since the 1980s and its reasonable to think that Japanese can build leading edge semis using TSMC know-how and ASML machines. Japan's re-focus in this area seems like a good long-term bet.
Good points. A lot of people who write Japan off miss the fact that its manufacturers moved up in the supply chains, focusing on B2B instead of people. So there's still a lot of know-how here. But I miss the customer-facing stuff.
The consumer-facing Japanese software would be the Playstation and the Switch, both respectable sized businesses and both fit into the model you speak often about- how Japan sells dreams. The other would be the IP owners who sell the 'software' that is anime- via Crunchyroll or Netflix, etc.
That Japan has the two largest console gaming platforms and has kept those positions for decades was not a foregone conclusion. Many have tried/are trying to take those positions from Sony/Nintendo to little success. If the a large portion of Japanese creators and software engineers are now building games software, it's enabled Japan to keep a key stronghold in gaming that most others envy and has only grown in influence.
Japan would be on the front lines of semiconductors, except that the US developed the essential technology (EUV lithography) and only licensed it to ASML because we were suspicious of Japan at the time. Canon/Nikon sell perfectly good… last-gen machines.
OpenAI and Google do have labs in Tokyo. The one startup I know (sakana.ai) is from people who left Google.
To develop your own LLM you need a few things: a lot of PhDs who speak English, good software people, high wages, a billion dollars, cheap electricity, and friendly enough copyright. I think Japan doesn't have enough of those things. The smallest country I know who've made a good one is France (Mistral).
The focus on robotics might have distracted them. Robotics turns out to be much harder than language (this is called Moravec's paradox) and far as I know nobody else is doing well there or even beating Japan.
And Japan did actually develop its own LLM last year, using the Fugaku supercomputer, which is no slouch spec-wise. But it was a much smaller model than any of the foreign rivals and didn't make much of a splash, perhaps because it doesn't seem to have been released publicly. The only really notable thing about it is that it was trained on Japanese-language data.
A detail to edit: In the second paragraph, you talked about 2002 as the year that ChatGPT and DALL-E came out... I think you mean 2020.
2022! Thanks for the catch. (That isn't the year ChatGPT was invented, but it is the year it exploded in popularity.)
Japan has not had a globally popular software app (outside of gaming) in decades. In the software realm, Japan is far, far behind still and I see no reasonably fast way for Japan to leapfrog like China did skipping the PC paradigm and focusing on mobile first.
Japan has kept a key spot in the semiconductor manufacturing supply chain since the 1980s and its reasonable to think that Japanese can build leading edge semis using TSMC know-how and ASML machines. Japan's re-focus in this area seems like a good long-term bet.
Good points. A lot of people who write Japan off miss the fact that its manufacturers moved up in the supply chains, focusing on B2B instead of people. So there's still a lot of know-how here. But I miss the customer-facing stuff.
The consumer-facing Japanese software would be the Playstation and the Switch, both respectable sized businesses and both fit into the model you speak often about- how Japan sells dreams. The other would be the IP owners who sell the 'software' that is anime- via Crunchyroll or Netflix, etc.
That Japan has the two largest console gaming platforms and has kept those positions for decades was not a foregone conclusion. Many have tried/are trying to take those positions from Sony/Nintendo to little success. If the a large portion of Japanese creators and software engineers are now building games software, it's enabled Japan to keep a key stronghold in gaming that most others envy and has only grown in influence.
Japan would be on the front lines of semiconductors, except that the US developed the essential technology (EUV lithography) and only licensed it to ASML because we were suspicious of Japan at the time. Canon/Nikon sell perfectly good… last-gen machines.
OpenAI and Google do have labs in Tokyo. The one startup I know (sakana.ai) is from people who left Google.
To develop your own LLM you need a few things: a lot of PhDs who speak English, good software people, high wages, a billion dollars, cheap electricity, and friendly enough copyright. I think Japan doesn't have enough of those things. The smallest country I know who've made a good one is France (Mistral).
The focus on robotics might have distracted them. Robotics turns out to be much harder than language (this is called Moravec's paradox) and far as I know nobody else is doing well there or even beating Japan.
Interesting you bring up PHDs, because there was an interesting post about that on Japan Economy Watch:
https://richardkatz.substack.com/p/how-does-the-dearth-of-phds-hinder
And Japan did actually develop its own LLM last year, using the Fugaku supercomputer, which is no slouch spec-wise. But it was a much smaller model than any of the foreign rivals and didn't make much of a splash, perhaps because it doesn't seem to have been released publicly. The only really notable thing about it is that it was trained on Japanese-language data.
https://www.fujitsu.com/global/about/resources/news/press-releases/2024/0510-01.html