23 sections now… and *maybe* it has enough structure now that this is going to get released as a post-a-day series that runs for a month instead of just an apocalyptically huge monolithic mess

8500 words, but that's after I just deleted my first section of about 800 words, which suggests that finally something like a shape of a final edit might be taking place

Hello if you are in the bay area over the weekend and are interested in technology and how it affects actual people, may I recommend popping up to Petaluma for @NorthBayPython? It's this weekend!

Talks and tickets: https://northbaypython.org/

(We'll waive ticket costs on request too, if you need it!)

You can measure your skill at DIY house projects by the ratio of visits to the hardware store against days of work:

- no visits: god tier
- less than one visit per day: excellent, you know what you’re doing
- 1 visit / day: average
- more than 1 visit / day: reconsider your life choices

As foretold by prophecy, LAPD used their "first responder" drones to spy on No Kings protests and anti-ICE protests: https://theintercept.com/2026/04/20/lapd-skydio-drone-surveillance-no-kings-protest-ice/

Only another 48 hours before you can no longer buy #NPBy 2026 tickets before the start of the conference.

But why not buy now? https://pretix.northbaypython.org/nbpy/nbpy-2026/

I need a graph of context window growth vs. resources like inference compute / RAM over time. Has anyone already made this?

And does the model's size itself put limits on context windows, or is that purely a feature of the hardware you're doing inference on?

Old habits die hard. Needed to choose randomly one of 18 items printed out on a piece of paper, with some exclusions. Went to #Python:

random.choice(list(set(range(1, 19)) - {5, 11}))

Are you 18+, neurodivergent, LGBTQIA+, and live in the USA? The PRIDE Study (pridestudy.org) would potentially love you to review our next annual questionnaire, from a neurodivergent perspective!

More details here:
https://stanforduniversity.qualtrics.com/jfe/form/SV_ermeCWynM8JBXz8

"AI" is going to make getting software engineering jobs harder over the next year... but not because "AI" will replace humans, but rather because of the interaction of accounting and the need to pacify stock market investors.

1. Megacorps have spent massively on AI datacenters. This is a capital expenditure which means the expense isn't booked immediately.
2. Instead, the expense gets turned into deprecation over the next few years, i.e. reported expenses are going to be much higher for the next few years.
3. Higher expenses shrink reported profits, so these companies will start doing layoffs.
4. So far there is no evidence of offsetting profits.

Notice this effect happens _even if all AI datacenter spending stops today_. It's baked in as depreciation.

So far we've had layoffs at Amazon, Meta, and Oracle (much more dire straits financially than the others). Today Microsoft is doing voluntary worker buyouts: https://www.bloomberg.com/news/articles/2026-04-23/microsoft-offers-voluntary-retirement-to-about-7-of-us-workers

I've seen rumors of Google doing buyouts too. And when these layoffs hit software engineers, that's going to mean a lot more competition for any open jobs. These companies have also hired a lot, and will be hiring less.

So that means more applicants for fewer jobs.

This week in "wait they weren't already a PSF fellow?!", congrats to @polkphebe for her well-deserved addition to the roster!

Bonus round!

When you die, your family will be charged $100 for every time you've ever honked your car horn. I cannot tell you how I know this, but please just understand with all sober certainty how very important it is that you never again honk your car horn.

brb changing my social media bio on every site to "guy who might one day write a blog post about LLMs"

I want to be clear that this is pretty divorced from my own views of likely scenarios, even likely nightmare scenarios, so they're not getting this from me.

If you're reading this and you're involved in LLM hype, I'd really like you to take a moment to pause and examine the fact that this is the commonly accepted peer wisdom in a 2nd-grade classroom right now. That their parents' generation is actively involved in betraying them to a distant and inscrutable machine-god of pure greed.

Talking to my 8-year-old about my upcoming Way Too Big Blog Post About genAI. I explain that my goal is to describe why I think it's bad, because it's really complicated and hard to understand, and so I don't blame people who disagree.

They ask, matter-of-factly, "Oh, are you trying to get people to stop using it because if they don't stop then by the time my generation grows up, nobody will have jobs any more and everyone will be poor because AI will just do everything and keep all the money?"

"How paranoid are you?"
*guy who is me who just built a web application to get notified when imaging satellites fly overhead*: "Just paranoid enough to be funny"

I know y'all are nerds (non-derogatory) but please do keep in mind that one of my goals with converging on a term to use consistently is making my writing easily comprehensible by a large audience. I am talking about the crud that we all have to talk about all the time now, and I want to be as congruent as possible with common usage without unnecessarily contributing to hype

Bonus round!

Style guide question: how to consistently refer to the umbrella of generative AI, including LLMs and diffusion image/audio generators?