Recent Posts

me, giving notice with nothing else lined up amid a sea of industry layoffs, and getting a job offer from a better company (not another fucking startup, lol) eight days later

Hey, maybe the robot uprising will begin there! We can only hope! 

16 years after this thread began with them, the Secret Service is using Boston Dynamics robot dogs to patrol Mar-A-Lago.

This floated across my feed today: How One Woman Became the Scapegoat for America’s Reading Crisis

tl;dr it's about Lucy Calkins, who I'd never heard of, but she's considered to be the main person responsible for schools de-emphasizing phonics over the last couple of decades (which is a thing I was vaguely aware of from talking to friends with school-age children) which in turn is likely a significant contributor to falling literacy rates.  Truly wild how much lasting damage a well-intentioned bad idea can cause. 

Also wild that Why Johnny Can't Read was published in goddamn 1955 and we apparently have to relearn the same exact lessons every couple of generations.   

There's no hyper growth in long-term thinking.

once upon a time the whole purpose of higher level languages was to eliminate boilerplate, and silently "auto-generate" it in lower level machine code.

Yup.  Something I think about a lot: if you have so much boilerplate that you need an AI to help you copy and paste it, you should be thinking about writing better frameworks/languages, not better copy/paste tools.

Unfortunately, designing better frameworks requires high level and long term thinking, and the industry as a whole hates those now.

I've seen AI work well as a sort of auto-complete in places where you're basically copying and pasting boilerplate; it's pretty good at figuring out via context what boilerplate you need, and suggesting it for you.  When I'm writing unit tests is mostly when I've found it somewhat useful.  It's still only right about 80% of the time, though, which makes it useful only insofar as I already knew what I wanted it to type and it's just saving me the keystrokes.

Is obscene. Copilot and ChatGPT used to generate boilerplate, when once upon a time the whole purpose of higher level languages was to eliminate boilerplate, and silently "auto-generate" it in lower level machine code.

Can be useful, but you have to know what you're looking at.

Your "it works as an autocomplete" is the entirety of what it actually IS. It is just a fancy auto-complete with the added benefit (for the oil companies) that it burns a barrel of oil for every query.

I've seen AI work well as a sort of auto-complete in places where you're basically copying and pasting boilerplate; it's pretty good at figuring out via context what boilerplate you need, and suggesting it for you.  When I'm writing unit tests is mostly when I've found it somewhat useful.  It's still only right about 80% of the time, though, which makes it useful only insofar as I already knew what I wanted it to type and it's just saving me the keystrokes.

It's just one step further down the ladder from outsourcing engineering work to Eastern Europe and/or India.  Companies that lean heavily on that stuff eventually find out that they get what they pay for.  (Which gets back to my personal lesson being to just avoid startups, because the key word there is "eventually" and any company that's focused entirely on the short term will absolutely do all that stupid shit.)

The AI coding is just really an insane thing, IMO. Granted, I think most of these GenAI/LLM's are utter useless trash, but for coding, it's probably even more so. These AI chatbot things are using next word prediction to try to sound like a human, right? Coders should instantly see the problem - code does not sound like human speech or language. It's full of very domain specific phrasings, definitions, shorthand, and contextual cues. While a GenAI bot whose data was very specifically culled, curated, and built using very domain specific training data COULD be very useful, that's not what these shitboxes are. Also, that kind of domain specific products aren't sexy enough to get the venture capital to be built in the first place, because their potential is not a hypergrowth market.

Also, the execs are going to use the GenAI as an excuse to hire either less engineers, or cheaper ones right out of college. They won't give a shit that the work product is trash, just that it costs less to build.