Reading Log

by Kurt Pan

But now, the model is splintering, and users are drifting toward smaller, slower, more private spaces, like group chats, Discord servers and federated microblogs — a billion little gardens.

While content proliferates, engagement is evaporating.

Psychological problems such as depression and anxiety increase the risk of cognitive impairment in older adults. But mechanisms on the effect of psychological disorder on cognitive function is inconclusive. Repetitive negative thinking (RNT) is a core symptom of a number of common psychological disorders and may be a modifiable process shared by many psychological risk factors that contribute to the development of cognitive impairment. RNT may increase the risk of cognitive impairment.

https://www.minizinc.org/

Lots of similar interview questions are this kind of mathematical optimization problem, where we have to find the maximum or minimum of a function corresponding to constraints. They're hard in programming languages because programming languages are too low-level. They are also exactly the problems that constraint solvers were designed to solve.

Most people do not care to find the truth, they care about what pumps their bags. There’s a much bigger market for truths that pump bags vs truths that don’t.

The best model of a programming AI is a compiler. You give it a prompt, which is “the code”, and it outputs a compiled version of that code. Sometimes you’ll use it interactively, giving updates to the prompt after it has returned code.

You are still doing the coding, you are just using a different programming language.

Do the hard work and build better programming languages, compilers, and libraries.

They’re still just video games, where the action-response feedback loop is provided by software, not the universe itself.

And where it looks like the budget has been going are the screen rooms. They occupy the huge central spaces on the main floor of the museum, and I’m sure a lot of time, money, and passion went into these things. But it’s misguided.

I believe museums exist to present the real thing for the visitor to experience with their own senses.

Now more than ever in history, kids need a break from the screens that all too many of them are sadly often plugged into by default, and connection to the real world instead.

We detrain ourselves out of the ability to access the quality of the information and turn it into actual knowledge. So called “digital natives” that they lack the critical and analytical thinking skills to evaluate the information they find on the internet. You need a trained brain to actually benefit from the internet. The advertised benefits of all these tools come with a specific hidden cost: Your ability to think.

The more you work in a superficial way, the more brittle the knowledge foundation in your mind will be on which you base your cognitive actions.

If you can’t produce a comprehensive answer with confidence and on the whim the second you read the question, you don’t have the sufficient background knowledge.

In knowledge work the bottleneck is not the external availability of information. It is the internal bandwidth of processing power which is determined by your innate abilities and the training status of your mind.

You have to remember EVERYTHING. Only then you can perform the cognitive tasks necessary to perform meaningful knowledge work. We can’t circumvent the work needed to train our minds.

With the introduction of the iPhone 17 lineup and iPhone Air, we’re excited to deliver Memory Integrity Enforcement: the industry’s first ever, comprehensive, always-on memory-safety protection covering key attack surfaces — including the kernel and over 70 userland processes — built on the Enhanced Memory Tagging Extension (EMTE) and supported by secure typed allocators and tag confidentiality protections.

You run out of memory on a classical computer.

Ringbauer’s team was building a quantum computer that used not qubits but qudits — each with five possible states. The extra possibilities allowed each particle to hold more information, often reducing the number of steps needed for a complex computation.

Concave surfaces push rays outside the glass; convex surfaces keep them inside.

A displacement map is simply an image where each pixel's color tells the browser how far it should find the actual pixel value from its current position.

This prototype distills Apple’s Liquid Glass into real‑time refraction plus a simple highlight.

https://www.modular.com/mojo

Mojo keeps Python's familiar syntax while adding performance and control. AI people do actually love a thing. It's called Python.

RSS entered the game as a humble widget on the experimental My Netscape Network portal.

All RSS had to do to weather ICE, Twitter, AI, and whatever comes next, was keep things simple and let users build their own feeds, filters, lists, and aggregators. Like email, it probably won’t make anyone a billion dollars or reshape entire industries. But it will always be wholly yours. And if that isn’t nice I don’t know what is.

One of the important things to take away from this is that every dependency could be malicious. We should take the time to understand the entire dependency tree of our programs, but we aren't given that time. At the end of the day, we still have to ship things.

I used Apple Configurator to turn my iPhone into a dumb phone. I can only access the apps and websites I want to use.

Using self-control here was like a hunter-gatherer using a club in open battle with a nuclear society. Social media companies spent centuries of human effort figuring out ways to get me to check my phone. If I wanted to win, I would need to use guerrilla methods.

My phone feels like a utility.

I also use them for brain-dumping. I find it hard to keep a journal, because I find it boring, but when you’re pretending to be writing to someone, it’s easier.

Just avoid asking questions that can’t be easily verified.

People have secrets and some secrets shouldn’t leave your computer.

“Chat Control” refers to a series of legislative proposals that would make it mandatory for all service providers (text messaging, email, social media, cloud storage, hosting services, etc.) to scan all communications and all files (including end-to-end encrypted ones), in order to supposedly detect whatever the government deems “abusive material.”

We cannot let authoritarians wear us down until we lose all our privacy rights. Our privacy rights are fundamental to so many other human rights, to civil liberties, to public safety, and to functioning democracies.

If you want to support your favorite authors: send then money. A dollar helps more then viewing ads ever would.

Intel’s current CPUs just are not stable. I am giving up on Intel for the coming years and have bought an AMD Ryzen 9950X3D CPU instead. I wanted the fastest AMD CPU (for desktops, not for servers), which currently is the Ryzen 9 9950X, but there is also the Ryzen 9 9950X3D, a variant with 3D V-Cache.

Today, we don't need to worry about hardware-accelerated compilation (hopefully), and we have better tools for refactoring (thanks, Claude). But with formatting, we regressed.

I use uv as my Python package manager and build tool. It’s all I need to install and manage dependencies.

Ruff combines isort, flake8, autoflake, and similar tools into a single command-line interface.

ty is a type checker for Python.

pytest is the most popular testing library for Python.

Pydantic is a data validation and settings management library for Python.

I use MkDocs for documentation and static generation of the website for the project.

I use FastAPI for building APIs.

Handling dependencies is a pain, but Dependabot makes it easier. It automatically checks for outdated dependencies and creates pull requests to update them.

There are three major downscaling algorithms: nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation. Each algorithm requires a different approach to perform an image scaling attack. Furthermore, these algorithms are implemented differently across libraries (e.g., Pillow, PyTorch, OpenCV, TensorFlow), with varying anti-aliasing, alignment, and kernel phases (in addition to distinct bugs that historically have plagued model performance). These differences also impact the techniques necessary for an image scaling attack. Therefore, exploiting production systems required us to fingerprint each system’s algorithm and implementation.

Enter your email to subscribe to updates.