A faster Easter date function
21 points by eatonphil
21 points by eatonphil
Nice work!
This bit broke my heart, though, for reasons that are difficult to articulate:
Although I beat the AI in this case, it’s a pretty mild victory as it was just a basic model with a moderate amount of compute. It’s sad to know that if frontier models were pointed at the problems I’ve been tackling in this blog series, they would probably unearth every last jewel of optimisation left in the blink of an eye. I hope that I have time to explore this space more before it becomes barren.
It's not that I think that AI is a scam. I fear rather the opposite: As AI improves, it's starting to take over more and more areas of human endeavor. We are gradually ceasing to be a species of thinkers, and slowly turning into a species that tries to outsource all cognition and creativity to something we feel is smarter (or at least harder working) that us, and then acting proud of work we did not do.
I think of Iain Banks and the Culture a lot: His universe contained human-level sentients, but it also contained Minds. It was a given that even the least of the Minds could out-think and out-create literally anything a human-level intelligence could do. Now, in Banks's novels, the Minds chose to essentially keep humans as pampered house pets. And the humans found reasonable ways to amuse themselves: board games, parties, art. But board games between a Mind and human weren't remotely interesting, because of course the Mind would win.
And yet, we have an increasing number of people who take their vague ideas to the baby version of a Mind, and ask it to do all the work, then act all proud as if they had much to do with it. The reality with each passing month is that the machine is doing more and more, and the human is increasingly superfluous. And... I've seen what social media has done to people, and I expect this to be 10x worse.
I don't mind the Opus 4.5-4.8 tier quite so much, nor local models like Qwen3.8 27B or Qwen3.8 Flash Next. They're still incomplete enough that they can make a decent "assistant" or a "virtual pair partner", if someone uses them that way. But more recent and larger models bother me at a very deep level.
Despite not caring in the least about date functions these posts have been some of the best technical writing I've read in recent years. The mod 2^n-1 trick (and its generalization) from a previous post caught me offguard, despite being very well acquainted with all sorts of modular reduction tricks and techniques.
Thanks!
I spend a lot of time working on the writing for the blog posts, so I'm glad it pays off.
It might sound strange or pompous, but writing these blog posts is starting to feel not too different to the writing of the optimised date functions. When writing code I'm trying to visualise how the processor can flow over the operations without stalling, and when writing the blog posts, I'm trying to do the same for the human brain over the text.
I will re-order the code many times to find what works best, and do the same for the sections of the blog post.
I'm refining my writing skills just in time for it to be redundant when the AI learns to do it well enough.