ganxel

Musician, enjoyer of snooker, retrocomputing and obscure but beautiful programming languages.

Last Notes

2026-10-29 15:15:37 UTC

- reply

and aeroplanes

2026-10-22 19:05:24 UTC

- reply

well, more head count in the form of yet another layer of middle management. That's always sure to help us mend that hole in the road faster...

2026-10-20 14:51:41 UTC

- reply

"because the /api/v1/chat/react-agent endpoint relies on a mock get_user_from_headers function that grants every caller the admin role"
Of course. Of course it does. Delightfully appropriate for something called "DB-GPT".

2026-10-19 12:06:29 UTC

- reply

time sync issues maybe? I had TOTP errors a while back, did a check on time.is and found my laptop was like 22 seconds slow.

2026-10-01 14:21:39 UTC

- reply

it sucks from the other POV too. We set up a small newspaper on Spaceship and used their SMTP for emails, but their own spam filter (jellyfish) blocked lots of them because it thinks Irish words must be spam (is bollox é sin dár ndóigh). When we got past that, it didn't matter because Spacemail was on multiple enterprise blocklists, so our shit bounced regularly.
Ended up having to switch to a dedicated transactional email service and migrate team emails to Gmail, so the big guys win again.

2026-09-18 21:47:11 UTC

- reply

at first I thought it was the spaced-repition flashcard program I used like 15-20 years ago... which would be a shame since the maintainer recently handed the project over to someone who disappeared: https://github.com/mnemosyne-proj/mnemosyne

2026-09-08 21:46:15 UTC

- reply

I had to come in for physical meetings in a crowded open-plan office where I could overhear people explaining to their "agent" why they couldn't go to an interview due to very personal medical conditions. And of course they could in return hear all my private shit. It was like going back to the early 1990s.

2026-08-28 21:52:18 UTC

- reply

that is such a bizarre inconsistency. I'm seeing people on Twitter accuse others of being "bigoted" against genAI, while happily insulting, pigeonholing and accusing those detractors of being NPCs. Like, what do they actually care abou?

2026-08-01 17:32:16 UTC

- reply

excellent -- another great post by the author here, on a topic that really grinds my gears: the fallacious analogies of "it's just a tool... you may as well object to the printing press or calculator".
https://www.psychologytoday.com/us/blog/the-algorithmic-mind/202607/ai-is-not-a-calculator-stop-saying-it-is

2026-07-28 14:12:30 UTC

- reply

"It may seem worse quality now, but it won't be soon (very soon in the case of AI)."
The LLM-maximalist bros have been saying that since ChatGPT first came out. I didn't buy it then and don't buy it now. LLMs are fundamentally limited by being a "plausible next token" text predictor, with no underlying reasoning or logical basis other than the emergent quality of its text prediction after overtraining/plagiarising almost everything humans have ever written.
This regurgitant mimicry of human-like writing has manipulated smart people (yourself likely included) into believing these mechanisms are more profound than they really are.

2026-07-28 11:50:34 UTC

- reply

"but realistically, like furniture or cars, etc., the non-handmade stuff will be better quality, more reliable, and faster to produce."
I don't think any of that is true except potentially the very last point, and then only for very small projects. As projects grow larger, the slop tax increases as the codebase becomes more and more of an unintelligible mess.