Xiluluke's Weekend Letter Vol. 11

This Week's Excerpts

The first principle of making money isn't selling your time. It's creating value.

The first principle of creating value isn't grinding. It's solving problems.

The first principle of solving problems isn't moving yourself to tears. It's finding real needs, testing fast, iterating quickly.

The first principle of iterating quickly isn't busywork. It's validating the model and replicating the system.

The first principle of replicating a system: leverage.

— Jike: Wang Zijun (Zima)

Life's five hard currencies: sleep, mental energy, muscle, metabolism, cash flow.

Three things every day: build models, build positions, build your body.

They map to: cognitive dividends, compounding assets, compounding health.

— Jike: Wang Zijun (Zima)

This Week's Flomo

1. Selected Works of Mao Zedong (Four Volumes) | "On Contradiction" | 2026-04-29

Quote: In China, there was the metaphysical thinking exemplified in the saying "Heaven changeth not, likewise the Tao changeth not," and it was supported by the decadent feudal ruling classes for a long time.

Thought: Marxism and materialist dialectics aren't right because they serve no class. Quite the opposite. Marxism openly declares its clear class nature and political nature, namely the revolutionary proletariat.
What makes Marxism and materialist dialectics right is their theory of knowledge and their view of development, which is exactly what Comrade Mao Zedong writes about here.
Knowledge has to come through constant practice, and reality rises in a spiral. This cycle of practice and upward spiral of development is the most perfect Golden Spin in philosophy.
That's exactly why Marxism and materialist dialectics can never be a dogma you can explain in a sentence or two. They develop through history along with human society, always changing.
This is also where Marxism's thoroughly critical and revolutionary nature shows.
That's what sets Marxism apart from most other philosophies.

Why Data Matters, and Changing My Strength Plan

After resting for a while, I officially started training again this week.

I did a review of my training over the past six months or so. Using the API built into the Xunji app, I easily exported all my past workout logs with Claude.

When I work out, I always log my weights, reps and sets in Xunji, every single time. Besides making them easy to look back on, this time having AI manage and analyze the data made me realize even more how important data is.

As I see it, data does three things: browsing, analysis and monitoring.

Browsing is data's first job. Flipping through set after set of neatly ordered numbers feels good, and it supports the next two layers.

Once you have enough data, you can analyze it. And long-running data across many dimensions also lets you keep a better eye on how the source of that data is doing.

Take my training. I exported all my workout data, didn't say anything else, and just asked the AI to analyze it in all sorts of ways based on my situation. I ended up with a really nice report.

Besides the overall analysis, it also tracked the weights for some of my most common lifts, like my bench press and squat below.

All in all, after looking at my data, I came to a pretty honest conclusion: in six months of training, I definitely got stronger. But given six months and how often I trained, the results fall a bit short.

Take my bench press. In six months it peaked at only 50 kg. Starting from a fairly low 20 kg, I think I should have reached at least 50 kg for a 12-rep max. I still can't do that. And since my training has been slack this year, my strength has obviously dropped again. Most days I train with 35 kg.

My squat did hit 100 kg once, but for someone my weight, that's nothing special. Because of my iliotibial band syndrome, I've only trained legs two or three times this year, and all with light weights.

My IT band syndrome has been getting better lately, so I think it's time to start training properly again.

So I've decided to drop my lazy training habits and start raising my strength ceiling.

In the end, I switched my program to the classic 5x5 strength program.

What's it about? It's actually simple. The program only has a few lifts: squat, bench press, barbell row, deadlift and barbell overhead press. The five lifts are split across two workouts, A and B.

Here's how I split them:

Strength workout A: squat, bench press, barbell row

Strength workout B: squat, barbell overhead press, deadlift

So what does 5x5 mean? Five sets of each lift, five reps per set.

Normally, for someone with some training under their belt, you can start this program at 75% of your 12-rep-max weight. What does that mean? Simply put, if 40 kg is the weight at which you hit failure at 12 bench press reps, then your 5x5 strength plan should start at 75% of 40 kg, which is 30 kg.

The core of the program is the A/B rotation, adding 2.5 kg each week.

Starting from Monday, if you make Monday, Wednesday and Friday your training days, you get an A-B-A rotation. You get two full rest days on Tuesday and Thursday, you can take the weekend off, and the next Monday starts with workout B.

After setting up the plan this week, I've already spent the last two days testing out both A and B, and set a reasonable starting weight for each lift.

Adding 2.5 kg a week doesn't look like much, and on the barbell plates it's next to nothing. But over the long run, in just one month the weight goes up by 10 kg. Train like that for a few months and my 5-rep max on every lift will go up by dozens of kilos, and so will my 12-rep max. That's the basic principle behind strength training.

Come to think of it, whether you're building muscle or losing fat, reshaping your body is fundamentally a long-term thing. It doesn't happen overnight.

And come to think of it, I haven't really put my heart into working out for the past six months. Mostly because IT band syndrome hit me hard, body and mind, and it wasn't until mid-March that I got my spirits back and started again.

This time, I hope I'll get to see myself make progress again.

Rethinking AI in My Writing

Lately I've used AI to write a few posts for my WeChat account. After taking a week off from posting, I still wasn't in the right headspace when I came back. Add a lot of stress at work, and I still wanted to keep my account updated as usual, so I started thinking about bringing AI into my writing.

In the end I had AI write about two or three posts, barely reviewed them, and published them.

My friend Yuehui saw them and messaged me privately to say my posts reeked of AI and you could tell at a glance. I didn't try to hide it. I told him straight out that AI wrote them.

He didn't say much more, just: "There's no end to giving in to AI. You'll probably end up using it more and more, and then you're in trouble."

I knew exactly how serious the problem was. I replied: "I'll make some changes."

Since that day, I haven't let AI write a single word for me.

Now that AI keeps getting more powerful, the more it gets involved, and the bigger its direct share in what you deliver, the less likely that is to be a good thing for people, especially people and work that need creativity and taste.

Take my recent experience vibe coding. More and more I feel it's no different from going to work. Every day I work and sleep alongside the AI, and yet apart from what it finally delivers, I have no idea what it's doing in the background.

And AI is getting more and more like a wheelchair, carrying you everywhere. I've not only given Codex full access, I've also let it control my folders freely and click around my desktop. Now I really don't even have to click through the testing myself.

It's seriously a wheelchair.

Of course, vibe coding is one thing. But when it comes to creative work, AI's access and interference should be kept as low as possible.

It's nice to have a friend like Yuehui. We don't talk much day to day, but a simple reminder like that at the right moment is like a sword of Damocles hanging over my head.

Wrong Punctuation and Typos Prove I'm Human.

Speaking of AI in writing, lately I've been working on a project that uses AI to generate text.

The more I dig in, the more one thing leaves me really, really speechless, though when I think it over, the AI isn't wrong either: punctuation.

How do I tell whether a post was written by AI these days? I look at whether its punctuation is perfectly proper and by the book.

Once a post keeps using “”, ——, ‘’ and the like outside of dialogue, I'll strongly suspect it was written by AI. Because I've never met a person who uses punctuation properly in every single spot.

That proper punctuation actually makes AI stick out among ordinary people. It's like delivering takeout in a bespoke suit. It's not that you can't, it's that you'll draw a lot of attention.

When I write, most of the time I use the usual punctuation: commas, periods, colons, quotation marks, exclamation points, question marks and so on. I rarely use things like single quotes or dashes. That's why I'm so unusually sensitive to them. Besides, I rarely see them in other people's writing either.

And of course, there are typos.

Some people are pretty careful about those. For a while I had AI check my posts for typos too. But now I've let it go completely. If it's wrong, it's wrong. Chinese readers have always been able to auto-correct text. Even with a typo, before you've even noticed it, your brain has already put the words back together correctly as you read.

In this day and age, less-than-proper punctuation and the occasional typo are actually the best proof that a living person wrote it.