Posts

I was listening to a podcast with Rick Rubin on his vibe coding book and he said some good stuff. The interview: Rick Rubin: Vibe Coding is the Punk Rock of Software The book: The Way of Code: The Timeless Art of Vibe Coding He said AI needs a point of view. That this is what we, as artists, supply. From the interview transcript: But the reason we go to the artists we go to or the writers we go to or the filmmakers we go to is for their point of view. The AI doesn’t have a point of view right. A point of view is what you tell it. What you tell it the point of view is to be. ...

A while back I showed how we can get a genetic algorithm in Python to run as fast then faster than a naive implementation in C. Here it is: Fast Genetic Algorithm in Python It was fun little project that mainly consisted of lots of timeit tests of different numpy functions and different ways of vectorizing the operations. The project was not very didactic. More of a progression. A show and tell. ...

I was thinking about DeepMinds AlphaEvolve from last month. It’s pretty cool. The results are cool. The overall structure is pretty straightforward-ish though. As in, it did not surprise me very much. AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms (announcement) AlphaEvolve: A coding agent for scientific and algorithmic discovery (paper, pdf) AlphaEvolve, Wikipedia Here’s how I’m thinking about it. It’s a GA with LLM operators. Something that many people have been trying, but have not unlocked this well, so far. ...

Back in grad school days, it was fun to pick NP-Complete optimization problems and have little competitions to see who could get the best solution (lowest cost) in a given time frame. One of the more popular competitions we had was made semi-formal called maxpath: Max Path Pic: I’ve always enjoyed optimization more than machine learning. I remember having coffee with Kaggle’s Anthony Goldbloom circa 2010 and saying that he should add competitions for optimization problems. It wasn’t a good fit. ...

Sometimes I don’t feel like working hard. I’ll procrastinate. Everyone does. I’ll fiddle with small stuff around the edges of the project. Bikeshedding. A trick I use to snap out of it is to “look back from the future”. I’ll pretend that I’m me but it’s the end of the year and I’m looking back at the achievements for the year. At all the projects completed. I will want to see the current project done, and more. If it’s not done, I’d be most disappointed. I’ll be asking hard questions, like: “Why the hell not? What were you doing exactly?” ...

I’ve been writing novellas for my kids for the last few months. The writing is fine. The stories, fine. Not mind blowing. The real trick is personalization. Telling stories about the kids as characters in places they know with our inside jokes. They are personalized stories. Written for them. This is key. It makes up for the so-so stories and writing. It’s a lot of work. It’s a heavy collaboration, a lot of careful specification by me of specific details to include. ...

I saw this on HN this morning: Show HN: I wrote a modern Command Line Handbook It’s currently number 3 with 63 comments and 245 upvotes. Nice! It’s for a book on how to use the command line on the big 3 OS. It’s a short, focused, basic handbook. I love those. I love making those. And I’m so happy that the tech community is still supporting them. And interesting in them. ...

I have a hobby as a quake archivist. It’s fun for two reasons: the nostalgia and the hunt (e.g. variable rewards). Many files I seek to archive are lost. They maybe in the deep web somewhere, or they may be in peoples personal archives, but unless I can surface them, they are essentially “lost”. Lost to the archives at least. I was thinking, soon the AI models will be good enough to generate list artefacts on demand. ...

Someone was asking me about investing the other day. I gave my normal spiel about not picking stocks and instead picking countries/economies and buying indices. I also said to stick with what you know. If you know property (she’s a real estate agent) then invest in property, you probably have some edge. The conversation flowed between investing ourselves and her kids starting to invest in stocks. I made the important observation that investment goals matter a lot and that the investment goals of hers will be different from her kids. ...

I was thinking about to early game modding. I made the most and best progress when I “just made stuff”. If in doubt, the second best thing to do would be to complete tutorials, e.g. coding tutorials, mapping tutorials, etc. The worst thing to do ever was to print/read manuals. There are two clear cases in my memory of doing this: Printing the DEU editor manual/guide (on my dot matrix printer) Print the Quake III Arena Shaders manual (in colour!). I never read them all the way through, and they never helped. ...