The web’s simple. Really, really simple. You send a text file over the wire, and you maybe get some key/value pairs back. That’s it. Everything’s a string, and the networking layer’s already been implemented for you.
Unfortunately, that’s far too simple to build a high-flying career on, so the industry collectively spent thirty years adding complexity to this simple model, and created the absolute clusterfuck of fragility, inconsistency, bloat, obsolescence, redundancy and instability we see before us today.
There was a stage, about fifteen years back, before cloud solutions were really mature, when engineers with too much time on their hands were all about scale. It was the problem (almost) nobody had, but (almost) everybody wanted. Scaling was sexy.
We all built “scalable platforms”, even when we only had three hundred impressions a day. Architecture astronauts would just cry “Scale!” when anyone questioned their latest whiteboard scribblings.
(“Mongo DB is web scale!“)
I’m sorry. We were bored, and building complex things is fun.
I’m serious, though - I think it’s like this because we let a bunch of really smart, really inexperienced children loose on what is, at core, a trivial problem. They complexified it because they were bored, and wanted recognition for their cleverness. In retrospect, we should have set my generation chewier problems, but unfortunately all the money was flowing towards web.
I wonder what the next “problem everybody wants” will be.
Quote
The rhyme-as-reason effect, or Eaton–Rosen phenomenon, is a cognitive bias whereupon a saying or aphorism is judged as more accurate or truthful when it is rewritten to rhyme.
For my John Pory project, I manually scanned a few hundred pages of microfiche. The scans are a bit wonky, and I wanted to tidy them up. An hour with a Well Known Chatbot™ and Kagi got me a quick and dirty Python script
It works by performing edge detection on the page, roughing it up a bit and looking for Hough lines. I know my images are almost straight, so I throw away any lines that are more than +/-5 degrees from the horizontal, average what’s left, and skew the image to fit. Uncomment the cv2.line() call to see the Hough lines. There’s still a bit of tilt to some of my images because “mean” is probably the wrong metric (I’m thinking longer lines should be given greater weight) but it’s good enough for my purposes.
I could have done it without the chatbot, but my Python’s rusty and my OpenCV is zilch so it would have taken most of a day. In fact I probably wouldn’t have bothered. It’s not coming for my job (yet), but it’s definitely a more effective lever. Hobbyist programming hasn’t been this much fun since the early 90s.