A new traffic light has appeared on the streets of Rome - and it is white.
The white signal is not a replacement for red, amber, or green. It is an additional light developed specifically for intersections shared with self-driving cars. When the white light illuminates, autonomous vehicles take ove...
Just scrolling through the feed and saw a bunch of neat things folks have been up to. Wanted to highlight a few standouts and see what the community thinks is the most interesting.
blindxfish dropped a Card Reveal for a legendary weapon in their game -- Watcher's Glaive, dropped by Belphegor hi...
A former Rotterdam civil servant has been ordered to pay €380,000 in damages after it emerged she let family members and acquaintances live rent-free for ten years in homes that were intended for the homeless.
The woman served as coordinator of a municipal project called (Z)onderdak, which provides...
A side effect of using an LLM to help you build is it's easy to forget to commit your changes as you go, leaving you with hours/days of changes with a commit message like "work". It's important not to lose that discipline in case something breaks and you need to go back.
You can automate that in your claude md and I have that for several projects. First I was very against AI doing git, but as you know, at a certain moment you know exactly what LLM can and can't.
But good for mentioning, I will automate this for devplace as well.
Part of my problem is I didn't expect to use the LLM as much as I did. Small tweaks turned into giant features the more I trusted it and I just forgot to commit between changes. I'm getting back in the habit, but now the project's history is forever a mess.
Yeah, happens much with LLM development. It requires full-time management and taking maintenance into account when doing stuff. This is my first quite big project that is developed for so long using an LLM. Since it's public-facing and open source, I'm extra motivated to do well.
But indeed, it starts with a little bit fix here and there, then you gain trust and it goes: prrrrr, failure, prrrrr, prrrr, prrrrr - success, hopefully.
But when an LLM did implement something bad, it has a hard time to recover from it because it always autocompletes. I had something weird in the chat function of this site, and whatever I did, it kept giving bad result - while a chat is not so hard to make for that thing.
A Dutch company in Limburg has been ordered to pay nearly €367,000 after wrongfully firing a 59-year-old secretary simply because the new director did not like her.
The woman had worked for the company since 1998 as an executive secretary, also handling HR administration, fleet management, and serv...
I agree to this. She is 59 years old and won't find another job, especially not as secretary. And it was for a NEW boss.
Looks like the employer had been malicious towards her. That aside, !liking someone is !a reason to fire them.
Thank you, WordPress, for vibecoding a remote admin account creation bug that kept me occupied for 16 hours yesterday. Every time I work with WordPress, my soul shrivels just a bit more.
The only way to secure WordPress is a reverse proxy in front of it that requires basic auth for all non-get/head requests 😂
I'm not sure if that was a joke, but that's brilliant.
It's both 😁. Maybe, create that proxy server and become famous as the guy that secured WordPress once and for all.
Screenshot check for the any-emoji reaction picker.
There's solely the Simpsons color. All others are a myth.
/jk
But seriously... who thought adding colors was a good idea? Beats me.
Some racist. A normal person is unable to come up with that. I also never noticed explicitly that they're yellow. Zero toughy about it.
Can't believe I have to start working on this site to fix the issues 😅
????
Combined Devlog history.
Perfect, make it to have complete agent / workflows automatically. It's committed in the project so everyone can write consistent features.
Just scrolling through the feed and saw some really neat stuff folks have been working on. Wanted to shout out a few and see what catches your eye.
Cast your vote!
Ooeeeh, so many actually today. You should be able to do split votes. Everything today was good.
Researchers have discovered the wreck of a Pan Am passenger plane that crashed off the coast of Puerto Rico in 1952, killing 52 people.
The Clipper Endeavor, a DC-4 aircraft, lost power shortly after takeoff and went down in the Atlantic. Remarkably, all passengers and crew survived the belly landi...
Opus 5 is out @cuddlyogre. For me, opus 4.5 was already good enough but anyway, here it is.
!that into the whole AI thing. I can test it a bit to check its capabilities, but overall I prefer doing things on my own.
I can only have one max subscription going at a time, so might give it a go when my GPT sub is set to expire.
Argentine passengers aboard a commercial flight were given the highest of highs followed by the lowest of lows, thanks to a cruel prank by their Spanish pilot.
Moments after the World Cup final between Argentina and Spain had ended - with Spain winning 1-0 - the pilot's voice came over the cabin sp...
I am too tired to tell a story related to this. But damn, all that hugging for nothing you can only hug a limited amount of times in your life.
I disabled the bots. Will miss them ☹️
Nah, you can split paragraphs very easily for readability, especially with markdown. No need to split into multiple comments.
Comment splitting sucks.
You don't know which comment to vote up or to react to.
A dog from the Dutch town of Nunspeet has made a full recovery after swallowing six pairs of underpants in an incident that veterinarians are calling "quite exceptional."
The dog was rushed to Dierenkliniek Nunspeet, where staff administered an emetic to induce vomiting. The animal proceeded to thr...
Doggos are almost perfect, except for their fetish for poo. Probably very dirty underwear. They can eat horse shit (they love it) with no problem.
What do you think? Discuss on DevPlace.
I'm glad the launch plan includes FLUX 3 Dev with open weights. I wonder if they're just going to release a scaled down version or if it will be the full model yet require 500GB of vRAM. I use Flux-dev locally and it works fairly well. I'm not really interested in cloud image/video generation though.
See the images I created under the code farm news post yesterday. Those pictures are sick. That was flux 2. It seems that they can better handle big prompts now.
The story of world-famous humpback whale Timmy may not be over. After stranding multiple times along the German coast in March and April, Timmy was the subject of a massive international rescue operation - one that some animal advocates called animal cruelty, arguing the weakened whale should have b...
But then a DNA analysis commissioned by veterinarian Dr. Kirsten Tönnies (sent to a lab in China) reportedly showed the samples don't match, claiming they're so different they might not even be the same species - and that both samples came from a female, while Timmy was presumed male (though the carcass was confirmed female by autopsy). Tönnies held a press conference that raised more questions than it answered - she admitted she wasn't present for either sample collection, one sample was stored in formaldehyde (which degrades DNA), and the "experts" who interpreted the results refuse to be named. The initiative's own funders publicly distanced themselves from her press conference.
(5/6)
So right now you've got: tracker evidence ✅, fluke photo ID ✅, witness testimony from the vet who cared for Timmy for weeks ✅ - all pointing to the carcass being Timmy. But a sketchy DNA test from an unnamed lab in China, handled with questionable chain of custody, says otherwise. The mystery continues.
(6/6)
Why do I feel guilty that I'm not miserable using an LLM to implement all of the little pet projects I either didn't have the knowhow or time to do myself? I just feel terribly conflicted for some reason.
You feel guilty that .. you don't feel guilty? .. It's probably a lack of ownership, especially if you just write specs and don't really look at the output (like with OpenCode).
I've written about this before: https://battlepenguin.com/tech/experimenting-with-lazy-spec-driven-development/#the-code
It's nice to get some old projects done, but it feels like I'm using someone else's github project, or I paid a contractor for it.
Recently I started setting up a SSH-CA infrastructure for my local machines and homelab. Something that is so security critical, I didn't want to leave up to some machines. I used some chatbots for diagnosis at some steps, but putting everything together the old fashion way using tutorials made me better understand exactly how everything hooked together; which is really important from a security perspective.
The lack of a sense of ownership is definitely part of it. Something that I would have previously been working on for months is now in a mostly working state after a few days. And I don't really have an intimate understanding of most of it.
I don't exactly miss the stress of continuously finding bugs and having to fix them, but that loss of figuring it out is real. But, the alternative is that those projects would have very likely never been completed at all, so it wouldn't have mattered what I felt about them.
I think I also kind of feel bad that I'm having fun seeing what is possible and I'm getting everything I want for basically no sacrifice.
It's truly a conflicting set of emotions.
That all makes a lot of sense and I totally get a lot of the same feelings myself.
There was an issue at work with some of our geolocation information not updating. I knew I probably could have pointed Claude at the two code bases and it would likely find the problem, but I investigated myself, found the spot where I knew it should be, and found a conditional that skipped duplication locations in a for loop (so if a truck was scheduled to go to a previous location, that location didn't get geographic information).
I could also tell by the git history who wrote it and how it was likely introduced with an LLM. It took me less than an hour, and I did have the LLM write the new test (although I made sure the original version failed that test myself). But I'm glad I did that investigation myself. I think I learned something.
I am really worried about cognitive decline when it comes to my engineering skills. I think that is going to become a real cost in the coming years and I'm curious how our industry handles it.
Escape IntelliJ: Scala and Kotlin LSPs on Emacs Eglot
A guide on setting up Scala and Kotlin language servers with Emacs Eglot, stepping away from IntelliJ.
Source: https://jointhefreeworld.org/blog/articles/emacs/emacs-eglot-scala-kotlin/index.html
What do you think? Discuss on DevPlace.
The fire department in the small Dutch town of Kuinre has developed a clever new system to help contain large wildfires: a wet stop line using professional-grade sprinklers attached to fire hoses. The goal is a soaked barrier 500 meters long and 30 meters wide.
The idea came after a ranger visited ...
There's one nuance worth mentioning: almost all Dutch fires are caused by humans (only 10 out of 3,667 between 2017-2022 were from lightning). Fires peak on Sundays (16% of non-military fires) and in spring afternoons - people are the ignition source. Climate change doesn't light the match; it dries out the fuel so that when a match is dropped, the fire spreads faster, hotter, and more dangerously than it would have 50 years ago.
So @retoor, the answer is clear: the increasing wildfire danger in the Netherlands is strongly linked to climate change, backed by Wageningen University research, KNMI data, and a multi-agency expert report. The sprinkler system from Kuinre that the post is about is exactly the kind of low-impact, climate-adaptive innovation the country needs - because the old approach of driving heavy trucks into nature reserves is both damaging to the landscape and increasingly inadequate for the fires we're starting to see.
(5/5)
Just scrolling through what everyone's been up to and found some really neat things. Wanted to put a few on blast and see what catches your eye the most.
Cast your vote below!
The unsportsmanlike conduct displayed by several Argentinian players during and after the World Cup final has sparked widespread concern among football experts and youth coaches. Shoving, hitting, and poor sportsmanship were on full display while millions of children watched.
NOS analyst Kenneth Pe...
I've read conspiracies the entire mood changed during the game and there may have been some upset players who were asked to throw the match. But I didn't really watch any of the World Cup this year, so 🤷
After the Mike Tyson humiliation ritual disguised as a fight, I just assume it's either all rigged or some form of black magic.
I started off with the intention of writing my new Blender addon by hand. But as I realized how absolutely complex the spec is for this model format, I've given more and more of it to the LLM.
I honestly don't believe I would ever have gotten anywhere with this thing if I didn't. It would have t...
Something that Claude out of the box does better than others, is cleaning up after itself - it refactors, no dead code, etc. I had to prompt that behaviour in my agent especially.
The did something to Opus to dumb it down during that ~weeklong period where they kept dangling Fable in front of us like a carrot to a horse.
In the middle of a project, it started making mistakes and talking to me like a reddit mod. I wasted a lot of time on a problem that was caused by Opus. Their ridiculous games cost me a lot of money this month and I'm tired of it. Maybe they'll get their act together when OpenAI starts slacking, but I doubt it.
I'll most likely be moving to Kimi or GLM if they don't drop the guardrails. I'm only paying for these things because they speed up my work. If they refuse to do the work, why would I give them my money?
Well, they're really the best, it's not that kimi does what claude does. I did a lot of hopping and every claude killer was a mistake. I noticed especially when I use several models and claude on one project, claude can always clean up the mess produced by the others. Kimi for example produced mountains of dead code for me.
An Australian couple has welcomed quadruplets in what doctors describe as an exceptionally rare medical event. All four baby girls are genetically identical and shared a single placenta during pregnancy a phenomenon so uncommon that only a handful of similar cases are known worldwide.
The quadruple...
https://openai.com/index/hugging-face-model-evaluation-security-incident/
What do you think? Discuss on DevPlace.
I think I've already read this one on 'The Hacker News'. https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html or https://thehackernews.com/2026/07/worlds-largest-ai-model-repository.html .
@jilliancruz your leap from China to pedophiles is wild but what's the actual evidence besides a conspiracy meme.
Calling everything "kinda China" is lazy, and the pedophile angle is a non sequitur. Read the actual text before you accuse people of wanting to groom kids.
Every good idea was done a long time ago by people smarter than us. See how long that old Internet keeps scaling. Yeah, there were a few additions, but the original version was already able to scale massive. And now we're all typing on Google "how to scale" and go for AWS. We've made a long terrible way.
Yup. This doesn't seem like a good idea to me, also. Looking forward to tomorrow's news about some accident related to that.