Mastodon Feed: Posts

db@social.lol ("David Bushell 🪿") wrote:

all modern software is either irredeemable garbage or teetering on the enshitty cliff, any good exceptions?

(wrong answers will be judged with silent scorn, right answers may unfurl the monkey's paw)

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

i mentioned this in passing yesterday, but in case anyone is wondering "where is the claude code longform piece" I sorta pivoted what I want to write about from "Claude code, what should be the best case scenario for large scale LLM code, is a dumpster fire according to this one snapshot in time" to "that and also it's getting worse not better."

So anyway, that is to say, it is getting worse not better.

jscalzi@threads.net ("John Scalzi") wrote:

Perhaps the only good thing to come out of the 250th anniversary celebration to date

https://apnews.com/article/mcdonalds-fried-apple-pie-america-250-962c7e82d7d5089bf63b2e7258f1c864

chipotle@mstdn.social ("Watts Martin") wrote:

I think that may be the first time I’ve sent an attachment in mail using Emacs’s mail client. Everything is so charmingly 2003, yet still modern. (I still haven’t set signatures, but that’s true for me in literally every other mail client I’ve used, so.)

jscalzi@threads.net ("John Scalzi") wrote:

None. I tend to work one project at a time, and I haven't started the new one yet.

RE: https://www.threads.com/@authorjaderowe/post/DZpUY8gjYs0

neatnik@social.lol ("Neatnik") wrote:

The idea was inspired by a blog post by @cabel that I read ~17 years ago and have thought about far too often since. 😅

http://www.cabel.name/2009/09/kashiwa-mystery-cafe.html

neatnik@social.lol ("Neatnik") wrote:

Doing a little email experiment, hopefully a fun one! If you send an email to previous@email.engineering, two things should happen:

1. You’ll receive a reply with the email that the previous person sent (the last email before yours).
2. The next person to send an email will receive the message that you sent.

Your identity isn’t shared (only the subject and body). Hopefully nobody will be a jerk and ruin this (the reply footer has instructions for reporting abuse).

Anyway, have fun!

jscalzi@threads.net ("John Scalzi") wrote:

I've never dropped acid but today I went through the disco carwash, which I figure is about the same

Attachments:

• gifv: Wild blobs of soap pulsing under LEDs

Boosted by glyph ("Glyph"):
petrillic@hachyderm.io ("Chris Petrilli") wrote:

All diodes are light-emitting if you try hard enough.

Boosted by glyph ("Glyph"):
catsalad@infosec.exchange ("Cat 🐈🥗 (D.Burch) :paw:⁠:paw:") wrote:

Every sign has a story

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

i'm pretty sure the bug here is that the spawnDepth or depth parameters are never actually used for anything or checked against a limit. i believe i have followed all assignments and accesses in 2.1.172, and checked for string constant accesses, but i am not finding any place where they are used.

it's important to remember that there isn't really such a thing as "controlling how agents are spawned" in claude code. Agents can spawn from the agent tool, but agents also have the agent tool. there is even a note in some new migration prompt that i hadn't seen before that says specifically that Opus 4.6 (the model used by the person raising the issue) loves to recursively spawn subagents. that was present before the "feature" of allowing recursive subagents was added in 2.1.172, because technically there wasn't really anything preventing agents from spawning subagents before either - the checkPermissions method from the AgentTool is hardcoded to "allow", or at least it was, i don't have time to align and apply the mapfile to new release versions today.

https://neuromatch.social/@jonny/116329076454434307

pzmyers@freethought.online ("pzmyers 🕷") wrote:

My daughter does science!

https://freethoughtblogs.com/pharyngula/2026/06/16/a-science-star/

dysfun@treehouse.systems ("gaytabase") wrote:

italicer, colorado

jwz wrote:

DNA Lounge Update, Wherein the Surveillance-Industrial Complex gains another foothold in SF
https://www.dnalounge.com/backstage/log/2026/06/16.html?utm%5Fsource=sp%5Fma

dysfun@treehouse.systems ("gaytabase") wrote:

my GP once said "dysfun, have you considered moving to colorado?"

dysfun@treehouse.systems ("gaytabase") wrote:

law of the piggy in the middle

jwz wrote:

Commodore (or some entity now calling itself Commodore) is releasing a flip phone. Smartphones are mindlessly seamless by design. Callback is built around a simple research-backed finding: remove features designed to pull you back in, & reintroduce...
https://jwz.org/b/yk8a

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

man i remember when cloning a repository didn't cost you $30 and you actually got the files instead of nothing.

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

look man i feel for this person but i am laughing and screaming in my lab right now. every line of this gets better (formatting mine)

Agents fetch individual repo files via HTTP instead of cloning: A poem

They now fetch individual files from GitHub via HTTP
one file at a time.
Each fetch is a separate tool call
carrying the ENTIRE CONVERSATION CONTEXT
hundreds of round-trips
each re-sending the full system prompt
and context
at Opus pricing

dysfun@treehouse.systems ("gaytabase") wrote:

i have noticed that people get upset about rules J and K. clearly we need an L rule that is even more upsetting

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

i'm gonna go ahead and bet money the problem is that both spawnDepth and depth are used interchangeably and there is always a fallback condition to 0 when the attribute can't be accessed. fallbacks are the LLMs absolute most favorite thing in the world.

edit: please note that this is a betting statement and i have not analyzed the code beyond an ast diff and am just browsing around. these are not confirmed RE findings.

Boosted by slightlyoff@toot.cafe ("Alex Russell"):
alcinnz@floss.social wrote:

"A lot of hype and noise is pressing us to extend the cowboy, wild-west phase of the software industry’s expansion. We should set that aside, and take ourselves seriously as a mature industry."

How building an HTML-first site doubled our users overnight - Alistair Davidson:
https://www.mohkohn.co.uk/writing/html-first/

dysfun@treehouse.systems ("gaytabase") wrote:

you are in dangerous territory having a voice clip conversation with me. it might be a 3 second clip, it might be a 3 minute clip and the only reason it couldn't be a 3 hour clip is because i can't possibly not accidentally cancel it somehow in 3 hours.

dysfun@treehouse.systems ("gaytabase") wrote:

signal's UI is so good i just recorded a voice clip three times because i managed to cancel it instead of sending it twice.

dysfun@treehouse.systems ("gaytabase") wrote:

water in video games is fake

omg, i have no words

dysfun@treehouse.systems ("gaytabase") wrote:

extreme ironing has been done on everest!

jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:

“Federal appropriations law requires the NSF to notify the House and Senate appropriations committees at least 30 days in advance of any planned decommissioning of agency owned facilities or assets valued at more than $2.5m. The House letter said no such notification had been transmitted.”

https://www.theguardian.com/us-news/2026/jun/16/ocean-monitoring-cuts-trump?CMP=Share%5FiOSApp%5FOther

dysfun@treehouse.systems ("gaytabase") wrote:

it was this craze where people would just iron clothes like on the side of a mountain or on a surfboard, on the back of a cow etc. https://www.youtube.com/watch?v=7fd6NBQ8ego

dysfun@treehouse.systems ("gaytabase") wrote:

holy shit i just remembered something.

does anyone else remember EXTREME IRONING?

Boosted by soatok@furry.engineer ("Soatok Dreamseeker"):
on_code@mastodon.gamedev.place ("Laurent 🫠") wrote:

Buying a game is a political act. Your money votes for the kind of industry you want. AAA is unsustainable: rising costs, AI threats, $1,000 consoles...
The future is smaller, bolder, and independent.
The talent fleeing big studios will go indie.
7/N