Week 14, 2025: Links and things

Following Week 13, 2025: Links and things, here comes Week 14, 25: Links and things! Although it is already the monday of Week 15.

Let’s kick off with some AI stuff:

• AI 2027 is a nauseating read: “We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.” Astral Codex Ten: Introducing AI 2027 introduces the project here, and I hope to also listen to – or maybe even watch – this podcast episode. Also see this post by Olle Häggström.
• Also, a report from Anthropic: Reasoning models don’t always say what they think. It feels like what this should be compared with in human psychology is not directly “dishonesty”, but more something like “priming” (whatever remains of that research).
• There are hidden memes inside of you – my friend John is booting up his “AI-salongen” series in English now in the form of “John & Julius Salon”. The concepts of hidden memes is interesting – I really like this intersection between human world and the AI world that John is exploring.

Then, some more mundane programming stuff:

• John Sundell was a blogger I read a lot back when I was doing Swift and iOS. Since I’m no longer following that scene very closely, I’ll have to admit I didn’t know he’s been away from blogging, but I’m glad he’s back: Modern URL Construction in Swift. Basically, it’s about how you with these macros they have these days in Swift can avoid the problem of certain constants being “optional” because you can prove it already at build time.

  This reminded me of a conversation I had with my coworker Andreas the other day. Andreas mentioned feeling conflicted about the desire for simplicity in code, on the one hand – and on the other hand, really liking type safety. I resonated a lot with that. Sometimes we type lovers have a tendency to overcomplicate stuff, in the name of type safety. And one such thing is reaching for code generation of various forms.

  It sure feels neat: now you can prove validity at build time, and only work with safe, type-validated, non-null objects at run time. But then all of a sudden you find yourself getting into all kinds of built-time complexity for, you might have to admit, not so much gain. Anyway, what’s kinda cool, then, is that languages such as Rust (famously), Swift (these days) and Zig are now building such code generation into the programming language itself.

• A deep dive into JVM start-up.

Finally, I bought a very cool domain name: ournewinsectoverlords.com. I’m thinking about what to do with it. Possibly just moving this blog there.