Movies

This week I went to see Spider-Man: Brand New Day.

After watching it, I got thinking about one question.

Does this film have a viewing threshold? Ever since Spider-Man came back to the Marvel universe, the threshold has clearly gone up, even though the basics of the character haven't changed: living in Queens, Aunt May's influence on him, and so on.

But as Spider-Man formed more and more ties with Tony Stark and others, his character became far more complicated than in the earlier Spider-Man series Sony made.

On one hand, young Spider-Man lost Aunt May and his girlfriend MJ, things that exist in the traditional Spider-Man story. On the other hand, as he ran into more and more people and events in the Marvel universe, those people and events had a powerful effect on his personal growth.

You could say Spider-Man in the Marvel storyline has long been different from the traditional Spider-Man of years ago.

His growth curve is more winding, and not until the end of Brand New Day did I see a mature Spider-Man.

So in my view, the theme of Spider-Man: Brand New Day is **"growing up."**

Lots of people online say it's an arthouse film with too many talky scenes, and that many people nearly fell asleep in the theater.

Thinking back on those comments after watching, I find them both frustrating and funny.

Frustrating, because audiences these days seem less and less able to appreciate dramatic scenes. They crave realistic effects, fierce conflict and dazzling fights.

Funny, because these viewers say they love Spider-Man, yet few really care about his growth and transformation.

Spider-Man is a screen character, but Tom Holland's Spider-Man alone has kept someone like me company through my whole youth. To me, Holland's Spider-Man is almost a living person, like a friend. I watch his story in the films and listen to him talk like I'm watching someone I've known for a long time.

So I don't think at all that Brand New Day has too many talky scenes. On the contrary, I think a film that wants to drive a character's growth needs a proper amount of them. And the pacing of Brand New Day, in my view, is flawless.

As for the people online saying the movie isn't good, I'd actually like to ask: do they know who the second Black Widow¹ is? Have they seen Thunderbolts? How many of them don't even know who the Punisher is? I can't say what the answers are, but I think that's where the so-called viewing threshold lies.

The Marvel Spider-Man's growth is a story that has spanned many years. Following and getting to know that growth is the viewing threshold. In these years even we viewers have changed from who we were back then. Many people don't even care about themselves, let alone a character on a screen.

But that's only putting Brand New Day in the context of the whole Marvel timeline. Can't an ordinary viewer judge this film purely as a film?

Of course they can, and even I can't say this film is perfect. It does have things it doesn't do well.

For me, the script is too easy to guess. Basically once you've seen the opening, you can predict most of the rest.

But I can't deny that when it comes to those comments online about Spider-Man having too many talky scenes, I can only say most of those people are probably outside the viewing threshold.

When the Punisher stands at the hospital window pretending to be Spider-Man, and Peter Parker calmly walks out of the hospital doors into the cheering crowd, stops, looks at the people shouting his name, and does the web-shooting gesture to a little kid, I think that's the best ending there could be.

If we're talking about the punch of the story, Brand New Day is far behind the last one, No Way Home. That film's idea of putting three generations of Spider-Man in the same frame was, for Spider-Man, no less than Avengers: Endgame. But sadly the creative team's creativity and imagination were too ordinary, and they turned Spider-Man into a childish character overflowing with saintly pity, wasting such a good idea.

So after watching Brand New Day, I'm glad Spider-Man didn't keep going down the wrong path.

Concert

This weekend I went to see the Chengdu stop of Phoenix Legend's "Auspicious" tour.

Twenty years ago, when smartphones weren't widespread yet, did Phoenix Legend ever think they'd still be this popular twenty years later? As a kid I grew up hearing "Above the Moon" and "Lotus Pond Moonlight" as phone ringtones, but back then I never imagined I'd one day sit in the crowd at their concert.

Phoenix Legend must be the most unique duo still doing concerts today. Their audience isn't niche at all. Men and women, young and old, they don't care. And they don't draw fans with looks either. If you had to say what draws people to them, it really is talent.

Come to think of it, even someone like me, a person born after 2000 who grew up in the countryside, knows their songs well. That alone shows their influence, and even more so in an era before online media took off.

Maybe because their work reaches so deep into the lower-tier markets, some people find their songs a bit too down-to-earth, or put bluntly, a bit tacky.

But I think that's exactly what shows how solid Phoenix Legend is.

Their songs are catchy to sing, and the lyrics are simple but evoke a lingering mood.

(A few lines from their hit songs.)

Most people buy Phoenix Legend concert tickets because of how much buzz they get online. But during and after the concert, I took a fresh look at the duo and their work.

Many people may not get their artistic style. But when it was over, all I wanted to say was that I'd never seen a concert where everyone could start singing along from the first line of the first song.

I'd also never heard such perfectly in-sync singing in a live venue. The singers never needed to lead the audience, because whatever the audience knew, they sang by themselves.

In this day and age, for Phoenix Legend to become one of the most popular acts on the concert circuit without relying on looks or a niche fan base shows that their art is accepted by most people.

Art that the people accept is truly good art.

That actually needs no proof at all. After all, back when listening to music on a phone was still a novelty, they'd already won the market's approval.

If there's another chance in the future, I wouldn't rule out going again.

AI

A rundown of my current AI subscriptions:

• SuperGrok Heavy
• Codex Pro 5X
• Kimi Allegretto
• DeepSeek API

July was a really anxious month. I subscribed to all of these in one go, and in fact now I have so many tokens I can't possibly use them all. But my thinking is stuck, so I have energy and nowhere to put it.

Seedraft was actually already a mature product, but two weeks ago I suddenly had a wild thought, a bold idea for change, and wanted to completely rebuild it. But once I got going, I found it's much harder than I imagined.

The old Seedraft was a workflow built on fixed steps and tons of prompts. After a lot of thinking, I realized a product like that hardly counts as an AI product, because users have too little freedom. For example, you can't have a deep discussion of the plot with it.

So I decided to turn Seedraft into a real agent app, keeping the workflow while giving users more freedom. That's my vision of creative software for the AI era.

I've talked through my thinking with a few friends, and they all basically agree with the big picture.

But building an agent with a fair amount of freedom on top of a fixed workflow is easier said than done for me. I know nothing about agent design and can only keep asking AI to clear up my thinking.

But after all that back and forth, my thinking was still a total mess.

Last week I saw this passage from Karpathy:²

I've found a useful trick when working with large language models (LLMs): a good long, rambling brain dump. Sometimes an LLM needs more "bits of information" to understand what you actually want, but you're too lazy to type it all out word by word. At times like that, I like to lean back, switch to voice input, and ramble for about 10 minutes, any old thing, pure stream of consciousness. Occasionally I'll say at the start, something like "switching to speech recognition, excuse any typos..." Sometimes I turn it into a few rounds of back-and-forth interview. But I've found LLMs are especially good at rebuilding order from these long, jumbled rambles, and what they echo back of your tangled thinking is often much clearer than what you first put out. The result is that your "mind meld" with the model gets smoother, and there's far less to correct from then on.

Following that idea, I did something: I made a phone call to Codex.

It was my first time calling Codex. It felt far too human. It deliberately mimics the pauses people make when thinking, so it really felt like chatting with a person.

I stumbled through my vision for the product and told it everything, then had it extract and abstract my product design. That way, Codex helped me clear up my thinking.

There's one more thing: whenever something about the product design comes to mind, say it out loud right away with voice input, as much and as detailed as possible.

After doing this for two or three days, I sent all of it to the AI and told it this was my vision for the product, but that I don't know many technical terms, and asked it to try to understand my ideas from what I'd said.

It worked really well.

Then, after several rounds of polishing and discussion between me and the AI, we ended up with a detailed product design document.

Now I find that when building a product, going back and forth with the AI over small details round after round isn't as good as talking things through with it at the start, letting it run for a few hours in one go, then reviewing twice. You can't tell the difference in quality, and you save a lot of your own energy.

So here's the lesson: every vision for a product has one point that's the core of the core.

Get that core right, and everything else falls into place. When you don't know what to do, write down your rawest, original idea for the product word for word, and hand it to the AI.

Also, as things stand, the best-value subscription is still GPT Pro, mainly because Tibo is so generous.

Next month I'll cancel SuperGrok and Kimi and keep only Codex and DeepSeek.

I have to admit that after the price cut, Codex's value is about the same as DeepSeek's...

Footnotes

1. The second Black Widow is Yelena Belova, who succeeds Natasha Romanoff as the new "Black Widow" in the Marvel Cinematic Universe. She appears in Black Widow, Thunderbolts and other works. ↩

2. Andrej Karpathy, a co-founder of OpenAI, former head of AI at Tesla, and a well-known deep learning researcher. ↩