Opening
Hi everyone. This is 服部明, one year into running my own company, coiai Inc. coiai is an IT company centered on technology, doing system development and product development. Every Monday on this show, I, the founder, share the reality of running a startup.
So, this week — well, I took a bit of a break last week, so this week I'm covering two weeks' worth of tech topics. And for the main segment, I wanted to share some things I've been genuinely wrestling with lately.
There were some big pieces of news this week — OpenAI shutting down Sora, and the whole controversy around ChatGPT's privacy issue. All right, let's get into it.
Tech News
The ChatGPT / Cloudflare Issue
First, this week's news. There was a story blowing up on Hacker News overseas about Cloudflare reading data while people type into ChatGPT.
Basically, while a user is typing their prompt, React is pre-reading it as they type, and that gets sent off to Cloudflare beforehand — Cloudflare being, roughly, an infrastructure service — so the communication was going out to Cloudflare before the user even finished typing.
Technically, this was for things like bot protection and preventing DDoS attacks, but since there was no explanation given to users, that's what stirred things up and made it a bit of a controversy.
OpenAI Shuts Down Sora
Next up, OpenAI's Sora. Its shutdown was announced pretty suddenly. There was a post about it on Twitter, and a notice on Sora's own homepage saying it was ending too. As I understand it, they said more detailed reasons for shutting it down would follow later.
My own guess is that the cost was simply enormous, and that they probably weren't able to establish a working business model around it — that's the kind of thing people have been pointing to.
Has anyone here ever actually run a video-generation AI locally? I've tried it myself, and it's incredibly heavy. Since it has to build the video frame by frame, it really is a huge load.
I'm not sure how many households even have a serious GPU, but even with something like a consumer RTX card, it's really heavy going, and the VRAM cap on what a typical consumer card has is usually 16GB or less.
So running a video-generation AI under those conditions puts a huge processing load on the machine, and I imagine that's exactly where the costs piled up.
It really drove home for me that no matter how technically fascinating something is, actually turning it into a viable business is a whole different, much harder problem.
Voyager 1's 69KB of Memory
Another story that was getting some buzz on Hacker News: Voyager. Voyager 1, launched back in 1977, is still sending data back to Earth today, running on 69 kilobytes of memory and an 8-track tape recorder.
69 kilobytes — that's absurdly small. About the size of an app icon image, maybe even smaller than that. And it's been sending data back from outside the solar system for 48 years on that. That's what had people excited about it.
The Rakuten AI Controversy — Can Japan Actually Build Its Own LLM?
And another story that blew up — the question of whether Japan can actually build a domestic LLM. Rakuten AI took some criticism over this. What it comes down to is that it's built on top of an open-source model — I believe it was based on DeepSeek.
On top of that, the fact that the MIT license file wasn't included when it was first announced added to the controversy.
Building an LLM from scratch takes an enormous amount of cost, so honestly, it probably would have been fine if they'd just said that clearly from the start.
And to be clear, tuning an existing model — which is exactly what Rakuten's AI here does — is a common practice, and there's nothing inherently wrong with it. But the fact that they marketed it as something like a "purely domestic" model, while receiving support from the Ministry of Economy, Trade and Industry, when it was actually a fine-tuned model — I think that's what really lit the fire here.
Someday I'd like to build one myself too — an LLM. Training takes a long time, so it's less that building one is technically impossible and more that training simply takes forever. The fine-grained tuning is probably brutally hard too, but I've been thinking I might try building a small-scale language model myself.
Giving Claude Code Long-Term Memory
This is another story I've been meaning to bring up — there was an article about how "giving Claude Code long-term memory changed the quality of brainstorming with it." What's interesting about this article is that, rather than keeping every single exchange with Claude as plain text, well...
Anyway, that's it for the tech news this week.
What I Did This Week — Getting Ready for the Startup Japan Expo
Next, I want to share what I actually did this week.
This week I was getting ready for the Startup Japan Expo happening in April — working on flyers, sales talk scripts, and also the design and prep for signage.
I'm doing all of this pretty much on my own, though I do talk things through with friends along the way — but the detailed, hands-on work is really just me, solo.
What Does Success Look Like for a Founder?
I'm not sure if this is true for every founder, but we're aiming to be a startup, more than just a small or mid-sized business. What's the difference? Well, we're also thinking about steadily putting food on the table, so in that sense we are a small business too, but I really do want to succeed on a big scale.
I think success as a founder can mean a lot of different things. The goal of simply being able to feed yourself is a wonderful goal. And wanting to be able to feed a lot of people alongside you is also a legitimate goal.
In my case, I want to work together with the people who've supported me up until now, and with friends too — rather than just being satisfied feeding myself alone, I'd rather be able to work alongside a lot more people. So it's a big dream, in that sense. I have a big goal, and that's what I consider success.
What counts as success for humanity as a whole, whatever — that's a separate question. But for me, right now, as a founder, that's roughly where success lands for me.
Breaking a Big Goal Down Into Small Tasks
Even when you're holding onto a big goal, what you actually do day to day is small stuff, right? Breaking big things down into small things — that's kind of what work actually is, I found myself thinking this week. Well, "this week," though honestly it's something I keep coming back to all the time.
In a sense, once I start having people under me, or outsourcing partners I can hand work to, handing over something vague and expecting them to just run with it would be really tough on them, right? The person receiving that kind of work would have a rough time with it too. So instead, breaking a big thing down into pieces small enough that someone can actually act on them — I think that's the real job of a manager, or a president.
On Our Mission Statement — Giving a Corporation a Soul
I think what I just talked about — having a dream I can share with everyone — is essentially what becomes a company's mission statement.
I've actually been wrestling with our mission statement right now, partly because the company has two founders — myself, and another person named Takemura, who's put in money as well. I'd like Takemura to commit a bit more to the day-to-day work, and I really do want to grow our company bigger. And once other people want to join us, we need some kind of compass to guide how we operate.
Right now, even deciding what kind of work we take on — it's not that we're fighting about it, but there are times I genuinely go back and forth on it. It's not like I want to take on literally any job at all. There are things I want to do and things I don't want to do. I don't want to do anything fraudulent, for instance. I have no interest in making money by deceiving people.
That line — what kind of work we take on — is exactly the kind of thing a mission statement can spell out.
A corporation really is what the word says — like a giant person made up of a gathering of people, I guess you could say. It's a bit like some kind of ghost or spirit, but it's a huge spirit built up out of everyone's intentions layered together. And if I want to give that spirit a soul, I think that's exactly what a mission statement becomes — the soul of the corporation.
Okay, and that's about it for this week.
Closing
We're about at time again this week. coiai Inc. works on web development, XR development, core business system development, and on-premises AI adoption support.
Our strength is getting directly out onto the ground and picking up what the people actually working there have to say. So rather than designing around what upper management assumes, we make a point of staying committed to the people on the ground, and our goal is always to build something genuinely usable — an app or a system that actually works — rather than something that just seems vaguely cool.
It's not about wanting to do something because it seems cool — it's about wanting to build something because it's genuinely convenient and there's genuine demand for it. So rather than working with companies that just want to hand everything off and forget about it, I'd love to work with companies that actually want to build something great together with us.
Consultations and quotes are always free, so please reach out about anything at all. See you again next Monday — I'd be happy if you'd subscribe to the channel and come back to listen again next week. Thank you.