The 'They Did It' Fallacy

I was just now listening to an argument that got me more and more interested as the argument went on. The author did a great job of first taking the opposite position, then building a case for their side, and then making a strong counterargument.

Essentially, the argument went something like this:

The Wi-Fi connection is unreliable.
Some people in the household say that the problem is that the router is too old.
However, after testing and reading reviews, it seems that the router is not the problem.
Therefore, the problem must be that the government is corrupt since they send dangerous radio waves to our house.

2025-02-08

On Interface Design

I have spent quite some hours building open source software. With that, a lot of things have gotten easier over time. For example, setting up tests, CI, documentation, and websites has gotten much easier. However, interface design somehow not. I already asked about it 3 years ago. If anything, it has gotten worse since I nowadays realize more how important it is to get the design right. Especially once libraries become more and more used, there is a real cost to introduce breaking changes.

2025-02-06

The transformrs crate: Interface for AI API providers

Recently, I was thinking again about where AI is going. I was wondering what I as a developer should, or should not, be building. I wrote a post about my thoughts and concluded two things:

Firstly, cloud-based AI gets much cheaper every year, namely about 90% cheaper. If your AI application costs $1 per day to run now, next year it'll cost 10 cents, and just a penny the year after that. And while the price drops, the models keep getting better.

Secondly, the best AI tools don't necessarily come from the big technology companies. Take Cursor AI for example. Microsoft had everything needed to make the best AI code editor - the most GPUs, the most popular code editor (Visual Studio), and years of AI experience. But a small startup built Cursor, which many developers now prefer. The same happened with DeepSeek. Google, Meta, Microsoft, and Amazon are all spending billions on developing the best models. But DeepSeek came out of nowhere and delivered great results. This isn't new. The same thing happened with Google in the early 2000s. Altavista was the biggest search engine until Google, a small newcomer, made something better.

2025-01-31

AI learning rate and some thoughts

Now that DeepSeek released their new model, I'm thinking again about where AI is heading. For some technologies such as electric cars or batteries, I think I have a reasonable idea of where they are heading thanks to learning curves. Due to price decreases, we will probably see more electric cars, drones, ships, and trucks. Oh, and of course stoves that can boil water in 40 seconds. For AI, I'm not sure. In this blog post, I'll try to estimate the learning curve for AI and see if I can use that to make predictions.

2025-01-27

Live reloading for any generated website

When generating a website (typically HTML and CSS files), it is often useful to have a live reload feature. This means that your browser will automatically reload the page when you make changes to the files via code. For example, say you write some code that generates a plot on a webpage, or that generates some WebAssembly module that is embedded in the page. In the past, I would use tools like webpack or try to manually establish a socket on the server and inject JavaScript in the page.

I recently found a much simpler solution. Just use Bash together with any server that can serve static files and injects live-reloading like live-server.

2025-01-25

Running code in blog posts is probably a bad idea

One of my favorite things to do is to automate as much as possible. So when I started writing blog posts, I thought it would be a good idea to run the code in my blog posts automatically. For example, I would add blogs with cool code and visualizations and then run this code upon each push to the repository via CI. I even made a package for it called PlutoStaticHTML.jl.

What PlutoStaticHTML.jl allows you to do is to write your blog posts in Pluto notebooks (Pluto.jl is like Jupyter notebooks but for the Julia language). Then, you can setup CI such that the code will be executed each time you push to the repository and the output will be embedded in the blog post.

2024-05-30

Battery Learning Curves

A few months ago, I spent some time on trying to predict the changes in batteries over time. My aim was to estimate when electric cars would finally be as cheap or cheaper than internal combustion engine cars. However, my approach in that blog post was extremely naive since I didn't know about learning curves. Simply put, learning curves are a real world phenomenon where a good or service becomes consistently cheaper over time. As shown in the learning curve link, solar modules have gotten 99.6% cheaper since 1976. From $100 to about $0.3, that's almost 4 orders of magnitude!

2024-05-29

Summary of Same as Ever by Morgan Housel

One of my favorite books this year has been Same as Ever: A Guide to What Never Changes by Morgan Housel. I've read some of his blog posts at Collab Fund before, so I already knew the book would be good. His writing style is simple on the surface, but at the same time surprisingly insightful. (Which is quite a feat, I wouldn't disagree if you say my academic writing has been the opposite.) Morgan also often comes up with new ideas that I've never heard before. For example, he has argued that the German army in the 1930s got so strong because they had their weapons taken away from them by the Treaty of Versailles. Thanks to the lack of old weapons, they built new and more advanced equipment from scratch.

2024-04-29

Some SaaS banned me, again

A few years ago, a good friend of mine happily moved to Microsoft Office 365. It was great, he said, because his files were synced across all his devices and he felt it was making him more productive. He even stored many gigabytes of personal pictures on OneDrive because the storage pricing was fair and it was easy to do. Then, he shared his picture collection with one person and got banned almost immediately. No OneDrive, no Word, no opportunity for backup, no nothing. Contacting customer support didn't help. If I remember correctly, Microsoft claimed he was possibly involved in terrorist activities. In truth, Microsoft probably had a problem with sharing pictures of children. But, he shared the pictures with his own wife. He shared pictures of his children with his wife. The Dutch tech site Tweakers wrote a long article about similar events with software as a service (SaaS) platforms in 2021, so this was not an one-off occurrence.

2024-04-25

The history of the ROC curve

The Receiver Operating Characteristic (ROC) curve is a well-known tool used to evaluate the performance of binary classifiers. Its history is clear. According to Wikipedia, it

"was first developed by electrical engineers and radar engineers during World War II for detecting enemy objects in battlefields, starting in 1941, which led to its name ('receiver operating characteristic')"

Another source also states that the ROC curve was first used during the second world war to distinguish between enemy targets, or just noise by radar receiver operators (utah.edu, 2007).

◀ prev

▶ next