Web dev at the end of the world, from Hveragerði, Iceland

Notes, Links, and Weeknotes (6 November 2023)

Work

This weeknote comes only a few days after the last, because last week was a little bit involved.

  • Worked on the print edition of Out of the Software Crisis.
  • Starting to see the finish line on the work on the import maps for testing course and finally started to get a vague idea for how to present the value proposition. Any and all feedback on the pitch page would be appreciated.
  • The course contains a surprising amount of Strong Opinions™ on code correctness for web projects. I suppose that shouldn’t be surprising considering it’s a course on testing, but what’s surprising is how much my opinions differ from what’s commonly held to be true in the field. Like I think Typescript is useless for asserting correctness or the lack of defects in your code. You know this as a user: pretty much every app you use has bugs and pretty much every one of them is written in Typescript these days. Typescript is great for developer ergonomics, but that mostly has the effect of making us faster at churning out buggy code. Well, it first makes so much slower because it adds so much boilerplate and nonsense. Then it gives us marginally nicer autocomplete, which I’m sure saves much more time than we waste on the rest.
  • Wondering whether I should drop the lesson on testing workers (web and service). I alternate between thinking that it’s an issue many people have difficulty with (esp. testing service workers) and that it’s a niche issue that affects very few, so I should spend that time on something else.
  • Realised that one of the reasons why I’m so fucking tired is that I’m doing everything on all my projects. Web design, typesetting, cover design, coding, writing – I even make the illustrations when those are needed. This is one of the reasons why I’m hoping people like the text-oriented web course format and that using the main site for promotion works. It means less work for me.
  • All of the above aside, since the course is nearing its launch I’m about to have time and availability for consulting work if anybody’s interested. Non-framework web dev is my speciality, although since I’ve been doing this for over twenty-five years, there honestly isn’t much in web dev I can’t help out with. Let me know if you think I can help.

Anyway, I’m about press the “fulfil these please, printing company” button on all the pre-orders for the print edition of the first book. It’ll be interesting to see how long shipping takes.

Pre-order the hard-cover edition of Out of the Software Crisis

I’m really excited about the course, myself, not in the least because I think the import maps workflow is a reset button for web development. I have no idea, though, whether I can convince others of the same.

“AI” and generative models

I haven’t commented much on the “AI” bubble lately because, honestly, I’d just be repeating myself.

  • Companies that fired a bunch of people and tried to replace them with “AI” discover that this leads to big mistakes. Surprise!
  • Yet another study shows that “AI” boosts perceived productivity and fluency while at the same time being filled with errors and mistakes. Oh, no! I never would have known!
  • Seeing every new study coming out show that “wow, the bias problem is even bigger than we thought! Whodathunkit?” I did. I thunk it.
  • Constant barrage of fact-free articles talking about how great this or that tool is.
  • Constantly seeing people who sought out your advice, ignored it, and now pretend that nobody warned them.
  • Or, worse, seeing them make tons of mistakes because of it, without ever noticing any of them themselves.
  • Even seeing other critics fall for the hype. (No, language models aren’t improving exponentially.)

Honestly, nothing has fundamentally changed since I published the book. The risks are the same. The dynamics and structure of the models are the same. There is some variation in how they are combined (hello multi-modal!), but there have been no fundamental improvements in the performance of the systems themselves.

It’s getting repetitive, and I don’t see the point in shouting from the rooftops the same warnings again and again every week, even if doing so would sell books and help pay the bills, and even though every non-AI entry to the newsletter seems to haemorrhage subscribers.

You can also find me on Mastodon and Bluesky