mg

Apple does not need to develop AI software itself in order to remain successful.

We more and more turn into cyborgs, wearing all kinds of processors and sensors on our bodies. And we need this hardware and the software that runs on it to be trustworthy.

Being the hardware producer and gatekeeper for trustworthy software to run on it is probably big enough of a market for Apple.

Even more so if their business of getting 15% to 30% of the revenue apps generate on the iPhone continues.

It has yet to be seen what type of company becomes most valuable in the AI stack. It could very well be that it does not operate LLMs. Similar to how the most valuable company in the hospitality industry does not operate hotels. It just operates a website on which you book hotels.

Looks like those are not open source?

I think I have seen people who do 3D modeling objects to later print them on 3D-printers use open source software in the browser. But I could be wrong.

Is there no good web based 3D editor that works in the browser?

Now that Chromium supports the File System Access API on desktop and mobile, I am building every software I need as HTML+JS+CSS.

Before that, handling files in the browser was cumbersome. You had to offer download and upload links for users to manage files. And handling whole directories was impossible.

But now web apps are like native software tools that you can use to edit and manage files on the file system.

And the bonus point is that what you build this way is fully functional on mobile too. Right away, without any changes. At least on Android.

Good TTS feels like it is something that should be natively built into every consumer device. So the user can decide if they want to read or listen to the text at hand.

I'm surprised that phone manufacturers do not include good TTS models in their browser APIs for example. So that websites can build good audio interfaces.

I for one would love to build a text editor that the user can use completely via audio. Text input might already be feasible via the "speak to type" feature, both Android and iOS offer.

But there seems to be no good way to output spoken text without doing round-trips to a server and generate the audio there.

The interface I would like would offer a way to talk to write and then commands like "Ok editor, read the last paragraph" or "Ok editor, delete the last sentence".

It could be cool to do writing this way while walking. Just with a headset connected to a phone that sits in one's pocket.

I'm confused about the rationale behind:

    |> fn($x) => array_column($x, 'tags')

Why is that inlined function necessary? Why not just

    |> array_column(..., 'tags')

?

I mean, I understand that it is because the way this operator was designed. But why?

I did not make any prediction. I made a calculation. About the market cap of MS if it grows into the AI era as fast as it grew into the Desktop era.

You seem to assume that the global wealth is static. If that is the case, how were those $500T created in the first place?

GDP is not static.

In the 80s, when Microsoft went public, the US GDP was under 5 Trillion.

It happens.

Life started with self-replicating molecules. And ramped up all the way to structures like the human body which consists not just of quintillions of molecules but of billions of quintillions of molecules.

Because, as can be seen in the chart, during the desktop era, Microsoft grew at 60% per year.

Growing at 60% per year for ten years is a factor of slightly over 100.

If Nadella grows Microsoft into the AI era as fast as Gates did in the Desktop era, it would become a 100 trillion dollar company within the next ten years:

https://www.linkedin.com/posts/marek-gibney_microsofts-story...

I just added it to Gnod Search for easy comparison with the other public LLMs out there:

https://www.gnod.com/search/ai

Interesting that they claim it outperforms O3, Grok-4 and Gemini-2.5-Pro for coding. I will test that over the coming days on my coding tasks.

Will this enable plain Debian on tablets like the the Microsoft Surface?

That would be awesome!

Douglas Hofstadter was my favorite author during my childhood and early youth. I’ve read every book he wrote from cover to cover.

He had a significant influence on my perspective on thinking, consciousness and AI.

I am more than surprised that he has made so little (and so negative) commentary on the current developments in the field of LLMs and other neural networks.

The biggest change I anticipate is larger cities.

When I ask people what they would like to change about their life, pretty much everyone mentions that their apartment is too small and too expensive.

Because everyone wants to live right in the center of a big city.

While at the same time, when I look at Google maps, big cities are tiny. I zoom out a bit and every big city is just a tiny speck in the middle of nowhere.

When construction becomes cheaper, it will be more compelling to build houses outside the city. And when driving in comfortable autonomous electric buses is available frequently and cheaply, living outside the city becomes more compelling.

I really don't know what to think of HDR.

I have yet to get any benefit out of it.

I disable it everywhere I can. In Instagram for example. When it is turned on (the default) every now and then I get some crazy glaring image in my feed that hurts.

Maybe it is because I don't play games? Is HDR useful anywhere outside of games?

No sound at all and the console shows lots of these messages:

    24:4 Uncaught TypeError: Cannot read properties of undefined (reading 'lang')
    at 24:4:132

Which seems to be this line:

    m.lang = e[(Math.random() * e.length) | 0]["lang"];

Another observation: You write:

    m = new SpeechSynthesisUtterance;
    m.text = "Hello";
    speechSynthesis.speak(m);

    Run that JS and your computer will speak to you!

But when I paste those 3 lines into the console, that also does not produce any sound here. I tried in Chromium and Firefox, both on Linux.

There is one showstopper, why I can't recommend Firefox to friends:

Mozilla's refusal to support the File System Access API.

With the File System Access API, we can finally build local first web applications.

I already wrote my own todo-list app and text editor and some other more specialized apps that work nicely in Chrome (On desktop and mobile). And I am in the process of writing a photo gallery too.

One can build workarounds for Firefox with old-fashioned download and upload buttons, but the user experience is miserable. Directory based tools like a photo gallery (for local photos) are not possible at all.

With the File System Access API, web apps feel just like local productivity apps.

> How do you figure?

Extrapolation. I see the progress I already made over the last years.

For small tasks where I can anticipate that AI will handle it well, I am already multiple times more efficient with AI than without.

The hard thing to tackle these days is larger, more architectural tasks. And there I also see progress.

Humans also benefit from a better codebase that is easier to understand. Just like AI. So the changes I make in this regard are universally good.

One reason is that humans have a strong tendency to optimize for the short term.

I witness it with my developer friends. Most of them try for 5 minutes to get AI to code something that takes them an hour. Then they are annoyed that the result is not good. They might try another 5 minutes, but then they write the code themselves.

My thinking is: Even if it takes me 2 hours to get AI to do something that would take me 1 hour it is worth it. Because during those 2 hours I will make my code base more understandable to help the AI cope with it. I will write better general prompts about how AI should code. Those will be useful beyond this single task. And I will get to know AI better and learn how to interact with it better. This process will probably lead to a situation where in a year, it will take me 30 minutes with AI to do a task that would have taken me an hour otherwise. A doubling of my productivity with just a year of work. Unbelievable.

I see very few other developers share this enthusiasm. They don't like putting a year of work into something so intangible.

You can think of them as the complex equivalents to normal integers.

Complex numbers have two components. If both are integers, the complex number is a Gaussian integer.

I'm not sure if every fractal can be expressed as an iterative formula f(z,c).

In 2012 I found a fractal by using a fundamentally different approach. It arises when you colorize the complex plane by giving each pixel a grey value that corresponds to the percentage of gaussian integers that it can divide:

https://www.gibney.org/does_anybody_know_this_fractal

You are right. I always wish for more specifics too when we talk about code here.

The library was https://mediabunny.dev/

Before I used my own proprietary code for media encoding/decoding. I also tested a WASM port of ffmpeg for a while.

Mediabunny's documentation might be fine for some developers, but personally I prefer a reference where I have a list of all functions and their specifications.

Yes, I understand the library much better now.

In the 90s a friend told me about the internet. And that he knows someone who is in a university and has access to it and can show us. An hour later, we were sitting in front of a computer in that university and watched his friend surfing the web. Clicking on links, receiving pages of text. Faster than one could read. In a nice layout. Even with images. And links to other pages. We were shocked. No printing, no shipping, no waiting. This was the future. It was inevitable.

Yesterday I wanted to rewrite a program to use a large library that would have required me to dive deep down into the documentation or read its code to tackle my use case. As a first try, I just copy+pasted the whole library and my whole program into GPT 4.1 and told it to rewrite it using the library. It succeeded at the first attempt. The rewrite itself was small enough that I could read all code changes in 15 minutes and make a few stylistic changes. Done. Hours of time saved. This is the future. It is inevitable.

PS: Most replies seem to compare my experience to experiences that the responders have with agentic coding, where the developer is iteratively changing the code by chatting with an LLM. I am not doing that. I use a "One prompt one file. No code edits." approach, which I describe here:

https://www.gibney.org/prompt_coding

How hard would it be to power all data centers via solar panels attached to batteries?

The area of solar panels needed to power data centers is .. maybe 100x the area of the data center?

Only a tiny percentage of the USA is covered by data centers. Maybe 100 million square meters? That would be something like 0.001%.

Then covering 0.1% of the USA with solar panels could power all data centers.

Cool. Are you running the model on your own server?

Over time, will humans turn all of planet Earth into an active system?

If so, the total addressable market of Nvidia might be pretty big.

Let's take the human body as a point of reference. The weight of the human brain makes up about 2% of the human body.

Earth weighs in at about 10^25 kg. 2% of that is about 10^22 kg.

All computer hardware on planet Earth weighs what? Maybe a billion computers times 10 kg? That would be 10^10 kg.

So we still would have to up that by a factor of 10^12.

Still 99.99999999% of the way to go.

I have written up a description here:

https://www.instagram.com/marekgibney/reel/DDezhSisD4J/

Thanks. Yes compression arteficts always become very visible. Especially in low-detail areas like the sky.

This reminds me a bit of a project I am currently doing where I swap adjacent pixels in the last image of short video sequences until it resembles the first image:

https://www.instagram.com/marekgibney/reel/DILksFYNSkE/