Life is a game. What matters is being on the journey.

I've run into a difficult problem. I haven't worked out a solution yet, but I'm not anxious. I even feel a little excited and happy.

Part of the problem in front of me is very similar to one I encountered two years ago, but things are different now.

In mid-2023, there were only GPT-3.5 and GPT-4. Large language models back then weren't multimodal, couldn't read links, and had limits on input and output length. Even feeding them material was a struggle; they would lose the thread. Basically, apart from a chat box, there was nothing.

I remember writing a business plan (BP) for AI-powered equipment diagnostics back then. It looked quite convincing, and the business logic was pretty elegant.
(I've forgotten why I wrote a BP, but later I came to think that writing one can sometimes be a rather good framework for thinking and a way to express ideas.)

Although I knew some of the teachers reading it might be investors, I hadn't written it to raise money, and I didn't think it could raise any. So the financial projections were all made up. Yet the teachers rated it very highly, and even asked me how the technical part would actually work and whether I could explain it in detail.

The problem was, given the form large language models took back then, I didn't know how to make it work either, pu~ I was simply reasoning from what they seemed capable of: it should be possible; people building applications just understood the business logic better.

But standing at this point in time, things are very different.

Over these past two-plus years, large AI models have changed a great deal. They have become multimodal, and their performance across different capabilities has increased severalfold. Technologies and tools built around them keep emerging, moving increasingly toward mature commercial products that can be used right away.

For example, all kinds of agent workflows, AI coding, workflows specifically supporting RAG (retrieval-augmented generation), general-purpose agent platforms like n8n with endlessly extendable functional nodes and connections to other systems, MCP (Model Context Protocol), and so on.

Two years ago vs. two years later: the problems we face may still be much the same, but we now have many more concrete tools and approaches in our minds and at our fingertips.

It feels like playing a game. As you go along, you keep encountering problems, but you're also picking up new equipment and skills along the way, helping you clear the next level, and then the next.

That's why I mentioned feeling excited and happy at the beginning. I know I have more capabilities and tools than before. I just haven't found the best solution yet, but I will.

It makes for a completely different state of mind.

Isn't this a kind of life game? What matters is that you're still on the journey.

I happened to come across this image of 雷军 (Lei Jun, Xiaomi's founder) on my phone; I don't remember when I saved it. I'll borrow his words as encouragement for us all, hei~