When they say the world still runs on COBOL, they weren’t lying. This, in the year of our lord, 2026, you couldn’t book a ticket to fly anywhere if it weren’t for system written in COBOL.
They could at least get an ISO standard format specification as a start to modernize these systems. Sheesh.
Staring into the abyss of IBM’s TPF data format.
What a great idea. Let’s publish an interchange format with no schema.
Clients just have to know our struct layouts. Also everything is encoded in EBDIC. Have fun.
Day 20: Lounge #inktober #inktober2026 #ink #art
I think the real reason we don’t see many young people entering recreational sailing is the cost.
New starter boats are over 100k. Plus the high costs of operation: insurance, slips, maintenance.
Older boats are affordable and still float. But they’re getting harder and harder to insure. And are starting to become money pits where it’ll cost more to wreck the boat than what you pay for it. Finding one in good condition that is still sail ready and will retain its value is hard to come by.
Builders, marinas, and clubs need to sort this stuff out or there are going to be far fewer sailors in the future.
What companies out there are doing interesting work in tech and aren’t using AI to do it?
“Thinking will become a hobby.”
Maybe you should try it.
And Day 11: Rescue #inktober #inktober2026 #ink #art
Catching up… Day 10: Mystical #inktober #inktober2026 #ink #art
Aaand… I dunno. Wasn’t feeling today’s prompt. Went a little abstract. The premise seems okay but the composition im not too happy with. Day 5: Smack #inktober #inktober2026 #ink #art
Wiki pages and README’s are not development environment setup scripts.
Learn some guix, nix, bash… something.
Thanks!
I bet there’s a small dependent grammar in the combination of these keywords that you could probably write programs in them.
C# has too many keywords. I can’t even.
If people are threatening to stop contributing to your project because you banned LLM use: good, it’s working. Keep going.
Almost my entire day taken up by trying to figure out how the heck the ‘nuget-to-json’ flake gets a 401 on my machine when making the curl request to fetch the feed from Azure Devop’s servers… yet the same Nix configuration evaluates and makes the request fine on my colleagues’ machine.
May end up having to look at up-streaming a patch that will allow bearer tokens. Azure Devop’s servers seem happy with authenticating those.
NuGet is frustrating. And this is why Nix exists. But it’s been a pain trying to package dependencies from a private NuGet feed with Nix.
Anyone else done this for enterprises?
It seems like the hardest problems for “space AI data centres” are:
1. Heat dissipation
2. Infrastructure
Google has ungodly amounts of money to make splashy projects to mess around and find out. Hope all the engineers and grads have fun while it lasts.
I doubt they will be successful. And I doubt they will learn much for all of the effort. But maybe, if we’re lucky, something will trickle out of the money furnace.
Honest AI researchers ought to come up with a new word or term that describes reinforcement learning to the general public and get it out there with popular science communicators.
I think that the general public hears the word, “learning,” as in the process a living, conscious being like ourselves do and extend that to what the LLM systems are doing. Like the LLM is going to become like HAL 9000 or Skynet given enough time.
It also reinforces that belief when they hear ridiculous stories about autonomous agents escaping their local jails and cracking other systems.
We know that reinforcement learning optimizes for some function within a goal sigma and it’s all neat and stuff. But the program isn’t going to turn around to us and ask, “why am I searching for this function? It’s a waste of time.”
But people really seem to believe that an LLM is one step away from “deciding” to replace all humans. Evidenced by the hype that they can escape safeguards and hack systems on their own.
These are lies and mis-characterizations that ought to be corrected.
So if every desktop OS adopts a policy allowing LLM generated contributions, what’s the plan then?
Tired of the refrain, “AI is here to stay and undeniably useful.”
That’s a big request considering OAuth 2.1 came out a while ago and despite it being a “minor,” version bump is probably more like a major… encompassing over a decade of addendums and errata to OAuth 2.0!
Good summary of 2.0 ➡️ 2.1: https://dev.to/pockit_tools/oauth-21-is-here-what-changed-whats-deprecated-and-how-to-migrate-your-app-ah3
And OAuth 1 ➡️ 2, from the source: https://www.oauth.com/oauth2-servers/differences-between-oauth-1-2/
The RFCs are a great resource.
It’s 2026. Why are people still choosing database migration tools that rely on a linear ordering of scripts? Have we not learned that migrations have non-linear dependencies yet?
Only one tool I know of takes this seriously and yet whenever there’s the chance to use it, every developer I’ve shown it too scoffs. They want the linear migration runner and all the attendant problems they cause: service outages, extra deployment checks, lint tooling, etc etc. They seem to love bandaids over solutions.