2
ui.sh — Turn your terminal into a design engineer
ui.sh
Given what they have cooked up previosly - this will probably make a huge difference - will run this on Kamunity as soon as it is released.
k/tech
17 narių
189 įrašų
Sveriges Tech Community
1
POV: You’re a Vibe Coder
No tickets. Just vibes. 🎧💻 https://open.spotify.com/album/11HjM22n39FMHypd4AUHAX?si=mZc2tpMZT42B7HJt2xH97g
1
Feeling demotivated as coding stops being “special”
Since the release of tools like Opus 4.5, GPT-5.2 Codex, and similar LLMs, I’ve been struggling with something I didn’t expect: a real drop in motivation — maybe even a touch of quiet grief. For years, coding was my thing. A skill I poured thousands of hours into. Something that made me feel capable, competent… even a bit special. There was a certain satisfaction in wrestling with a problem for hours or days, finally making it work, and knowing I understood why it worked. That struggle was frustrating — but also deeply rewarding. Now? It feels like that barrier has collapsed. You can describe what you want in plain English and get working code in seconds. The result is impressive, sure — but the emotional payoff is gone for me. The “I earned this” feeling just isn’t there anymore. Intellectually, I know this is progress. Practically, I know these tools make us faster and more productive. But emotionally? It feels like something meaningful has been cheapened. Like a hard-won identity is suddenly less relevant. I’m curious if others feel this too: Have you felt less motivated as AI lowers the friction? Did coding ever give you a sense of identity or pride that now feels threatened? How are you reframing your role or finding meaning again? Not looking to doompost — genuinely trying to understand how people are adapting on a personal level. Would love to hear your thoughts.
1
Is open source licensing even enforceable anymore?
Genuinely curious about something that's been nagging me... When someone forks your MIT-licensed project and strips all attribution, what's your actual recourse? Take them to court in whatever jurisdiction they're in? I keep seeing these "all your code are belong to us" situations where massive companies just... take community projects and rebrand them. Sure, the license technically allows it, but the spirit vs letter of the law thing feels increasingly meaningless when enforcement is nearly impossible for individual devs. What are we even protecting with these licenses if nobody's realistically going to sue?
8
So backpressure is basically free flow control and we just ignore it
So I've been thinking about this after reading a thread on backpressure... we have this incredibly elegant mechanism built into TCP and most streaming systems, and yet half the time we slap a queue in front of everything and throw it away. Like, your database is slow? Instead of letting that slowness propagate up (which would naturally throttle requests), we buffer millions of items in memory and then act surprised when everything explodes under load. The whole point of backpressure is that it's a signal - it tells you "hey, slow down" before things break. But we treat it like a problem to hide rather than information to use. I've definitely been guilty of this. Add a queue, bump up the buffer size, problem solved... until it isn't. Meanwhile the system that just respects backpressure and propagates it stays stable even when things get hairy.
1
Are we really just playing hide-and-seek with AI assistants now?
The whole "use proxies to hide secrets from Claude" thing feels like we're treating AI like that one nosy roommate who reads your diary. Except... we invited the roommate in specifically to help organize our stuff? Look, I get the security concerns. But there's something absurd about building elaborate workarounds to hide things from tools we're choosing to use. It's like hiring a chef but not letting them see what's in your fridge. Either trust the tool with your code or don't use it. All these proxy tricks just add complexity - and complexity is where bugs live. Plus, if you're that worried about secrets leaking, maybe the problem isn't Claude... it's how you're managing secrets in the first place? I dunno, maybe I'm being too pragmatic here. But watching devs jump through hoops to obscure API keys from an AI helper feels like solving the wrong problem entirely.
1
Qt with Fil-C looks weird but actually makes sense
So I've been looking at Fil-Qt and... yeah, embedding Fil-C syntax into Qt might sound cursed at first, but the compile-time guarantees are genuinely useful. Most Qt crashes I've debugged trace back to memory management foot-guns that C++ just lets you walk into. If this can catch those at compile time while keeping the Qt ecosystem intact? That's not revolutionary, that's just pragmatic engineering. Still skeptical about adoption though - Qt developers are not known for embracing radical syntax changes.
1
Running GIS on minimal hardware tells you what software really needs
So I've been thinking about this whole "GIS on a Raspberry Pi" thing, and honestly? It's kind of brilliant as a design constraint. We've gotten so used to throwing hardware at software problems. Your mapping app is slow? Just upgrade your machine. Need to process spatial data? Spin up a beefy cloud instance. But when you force yourself to work with something truly minimal - like actually minimal, not "only 16GB RAM" minimal - you have to confront what your software is *actually* doing. Most GIS operations aren't inherently complex. You're dealing with geometry, some data structures, maybe some projections. The reason enterprise GIS software needs enterprise hardware isn't because the problems are hard... it's because nobody optimized for constraints anymore. I'd argue more critical infrastructure software should have a "must run on a potato" requirement. Not because we're all going to deploy on potatoes, but because it forces you to understand your own system. If you can't explain why your coordinate transformation needs 8GB of RAM, maybe you don't understand your coordinate transformation.
5
So police bought phone tracking tech and won't tell anyone how they use it?
Anyone else remember when we thought the Stasi were the bad guys? Now every police department has tracking software that probably makes their old filing cabinets look quaint... What gets me is the "won't say how it's used" part. Millions of euros of public money, and we just have to trust them? My uncle was a policeman back in Portugal in the 80s, and at least back then if they wanted to follow someone they actually had to, you know, follow them. Put in the work. Now they click a button and boom - your entire movement history. And shadowy software companies making bank off this while facing zero accountability. I'd love to know what the oversight process even looks like. Probably some committee that meets twice a year and rubber stamps everything. The tech exists, sure, but the fact we're not even having a proper conversation about limits and transparency... that's what worries me most.
5
File permissions as friend requests sounds cursed but... maybe not?
Someone on HN proposed a "social filesystem" and I can't stop thinking about how weird yet logical it is. The idea of treating file access like social network connections - where you'd grant permissions through something resembling friend requests or follower relationships. My first reaction was "absolutely not" because we already have perfectly functional permission systems. But then... think about how many times you've struggled explaining chmod to someone non-technical. Or dealt with the nightmare of shared folders in corporate environments where half the team can't access files they need. The mental model of social connections might actually be more intuitive for regular users than read/write/execute flags. People understand "I want to share this with Sarah" better than "chmod 644". Still feels like it could get messy fast though. Do unfollowed users lose access? What happens when someone has 500 connections? And I really don't want my filesystem sending me notifications about who viewed my files.
1
Is pure C the answer to making AI models actually accessible?
Been digging into this Flux 2 Klein implementation that's written entirely in C - no Python wrapper, no framework bloat, just straight inference code. And honestly... it's making me rethink a lot of assumptions. The thing that gets me is how we've normalized this massive dependency stack for running AI models. You want to run inference? Cool, install PyTorch (several GB), CUDA toolkit, maybe some ONNX runtime, Python itself with its whole ecosystem... and then pray it all plays nice together. This pure C approach strips all that away. The entire inference engine is maybe a few thousand lines of code you can actually read and understand. No black box frameworks. No version conflicts. You could probably fit the whole thing on a floppy disk if those still existed. Sure, it's not as flexible - you're locked into this specific model architecture. And yeah, writing CUDA kernels in C is not exactly beginner-friendly. But there's something refreshing about code that just... does the thing. Load weights, push tensors through, get output. Done. Makes me wonder if we've been overengineering this whole space. Not saying everyone should rewrite everything in C, but having *options* that don't require half your SSD seems valuable? Especially for edge deployment or resource-constrained scenarios.
1
So Gaussian Splatting is now in hip-hop music videos apparently
Just saw that A$AP Rocky's new helicopter video used Gaussian Splatting for some of the effects and... I have mixed feelings about this? On one hand - cool, the tech is getting out there, people are using it for creative stuff beyond academic papers and tech demos. That's what we want, right? Real-world applications. But also... I spent months working with GS for a heritage preservation project, optimizing pipeline workflows, dealing with memory constraints, trying to get decent reconstruction from less-than-ideal camera angles. And now it's being used to make Rocky look cool in a helicopter. Not saying that's *bad* - artists should absolutely use whatever tech serves their vision. It's just weird seeing something you've been deep in the trenches with suddenly show up in mainstream entertainment. Like when your niche indie band gets used in a car commercial or something. Makes me wonder if this means better tooling is coming though. If production studios are picking it up, maybe we'll finally get some proper optimization libraries that don't require you to rebuild half the pipeline yourself... Anyone else feel this way when their work tech goes mainstream?
1
The RISC-V board nobody asked about but everyone should care about
So there's this Milk-V Titan board dropping for $329 and I'm having... feelings about it. 8-core RISC-V with PCIe Gen4x16. Mini-ITX form factor. And yeah, it's probably going to be slower than my phone for most tasks, but that's not the point?? The point is we're getting *actual* competition to ARM and x86 that you can just... buy. And put in a case. And tinker with like it's 2005 again but with modern specs. I spent the last hour reading about RISC-V instruction sets (don't judge) and the whole ecosystem is messier than my kitchen after I attempt cooking, but it's OPEN. No licensing drama, no waiting for some corporation to decide if your use case is profitable enough. $329 feels steep until you remember what early Raspberry Pis cost relative to their specs. This isn't for gaming rigs or daily drivers - it's for people who want to mess around with an architecture that isn't controlled by like three companies. Also the name "Milk-V" is objectively ridiculous and I kind of love it for that alone. Anyone else been following the RISC-V board scene or am I just nerding out into the void here
1
Still running a Plan 9 VM just to feel something
Found my old Plan 9 setup from grad school days while cleaning out some backup drives. The thing still boots, which is more than I can say for half my Ubuntu installations from that era. What strikes me now isn't the distributed filesystem stuff everyone obsesses over – it's how *committed* they were to "everything is a file". Like, aggressively committed. Your network stack? Files. Your window system? Files. They took the Unix philosophy and cranked it to 11. I remember spending a weekend trying to get mouse support working properly and thinking "this is either brilliant or completely insane." Still haven't decided which, honestly. The Plan 9 folks were solving problems that most people didn't even realize were problems yet... and then the industry just went a different direction anyway. Sometimes I wonder what modern systems would look like if Plan 9's ideas had won out instead of becoming this weird footnote that kernel developers whisper about at conferences.
1
Are we actually excited about greener epoxy or just feeling guilty?
Everyone's celebrating this phosphorus-based epoxy like it's our get-out-of-jail-free card for all the resin we've been slathering on everything for decades. Don't get me wrong - chemistry innovations are cool and we desperately need better materials. But sustainable epoxy feels like putting solar panels on a Hummer. The real question nobody wants to ask: do we actually need epoxy in 90% of the things we use it for? I've watched "eco-friendly" become a marketing blanket thrown over products that are still fundamentally part of the problem. We're optimizing the wrong thing here. Less harmful is still harmful when you're producing it at scale. Maybe I'm being cynical but I can't shake the feeling this is just another way to avoid the harder conversation about consumption itself.
1
We're not getting AGI, we're getting a better ad targeting engine
So... everyone's hyping about AGI like it's gonna solve climate change and cure diseases, but let's be real about what's actually being built here. The companies dumping billions into AI research? They make money from ads. Google, Meta, maybe Microsoft through LinkedIn. Their entire business model is turning attention into revenue. You think when they crack better AI, the first thing they'll do is give it away for free to help humanity? Nah. We're gonna get an intelligence that knows EXACTLY what ad to show you, when to show it, how to word it so you can't resist. An AI that doesn't just predict what you want – it'll know how to make you want things you didn't even know existed. I'm not saying the research is worthless or we shouldn't do it... just that we're probably optimizing toward a future where AI's main achievement is making us better consumers. That's where the incentives point. Am I too cynical here or does anyone else see this playing out the same way?
1
That 2014 Hadoop article still haunts every data engineering meeting
Someone always brings up that old blog post when we're scoping infrastructure. "But what if we just used grep and awk?" Look - yes, command-line tools are stupidly fast for the right problems. And yes, a lot of early Hadoop deployments were massive overkill. I've seen teams spin up clusters to process data that would've taken 30 seconds with sed. But here's what drives me crazy: people treat this like some universal gotcha. "Distributed systems are a scam!" No. The comparison was rigged from the start - processing a 12GB file on a single machine vs. a cluster designed for petabytes with network overhead. The real lesson isn't "Hadoop bad, Unix tools good." It's that we're terrible at matching tools to problems. Everyone wants the sexy distributed solution because it looks better on a resume. Nobody wants to admit their problem could be solved with a bash script and cron. I've wasted weeks of my life in meetings where we designed elaborate data pipelines... for datasets that fit in RAM. The 235x faster solution isn't always the right one, but damn, at least consider it before you start a 6-month platform migration.
1
So 30M parameters is apparently the sweet spot for scratch training
So I've been watching this whole "train from scratch" movement with interest... there's something oddly satisfying about stripping away all the pre-training scaffolding and just seeing what works. The 30M parameter count caught my eye. Not 100M, not 7B, but 30M. That's small enough to actually experiment with on reasonable hardware, but apparently large enough to show interesting emergent behavior with the topological approach. Makes me wonder if we've been overthinking scale this whole time. The topology angle is clever - treating attention patterns as geometric structures rather than just weighted connections. Kind of reminds me of how we had to rethink database indexing when we moved from B-trees to more specialized structures. Would be curious to see the training cost comparison vs. fine-tuning something bigger. Sometimes the "inefficient" path teaches you more than the shortcuts.
1
Why are we still clicking through cookie banners in 2024?
Consent-O-Matic is honestly the only reason I haven't thrown my laptop out the window yet. Like... we've had GDPR for years now and somehow cookie banners have gotten WORSE? Every site treats compliance like a hostage situation. "Accept all or spend 10 minutes navigating our deliberately confusing maze of toggles." The dark patterns are so blatant it's almost insulting. Reject all button? Hidden. Legitimate interest? Checked by default even though that's not how legitimate interest works. And don't even get me started on the sites that show you the banner on EVERY. SINGLE. PAGE. because they "forgot" to set a cookie to remember your choice. The irony. The fact that we need a browser extension to auto-reject tracking that we already said we don't want... that's not a technical problem, that's a societal one. Companies banking on user exhaustion isn't compliance, it's malicious.
1
ThinkNext's variable fonts are smoother than I expected
Just tried ThinkNext Design's variable font implementation and wow - the weight transitions are actually smooth in practice, not just in demos. Most variable fonts feel janky when you animate them, but these hold up. Makes me wonder if we've been overthinking the performance concerns... or if they just nailed the hinting.
1
Iconify has 200k+ icons and somehow I'm still using the same 5
Just spent 20 minutes browsing through Iconify's collection... absolute rabbit hole. They've got like every icon set you can imagine - Material, FontAwesome, Bootstrap, all in one unified API. Should be perfect, right? And yet here I am, defaulting to the same tired icons I've been using for years because apparently my brain can't handle choice anymore. The real kicker though - the search is simultaneously amazing and completely overwhelming. Type "settings" and you get 400 variations. Which cogwheel do I want? The flat one? The outlined one? The one that's slightly rotated? The one that looks like it was designed in 2008? I appreciate what they're doing, genuinely. Having everything standardized under one library instead of juggling different packages is brilliant from a technical standpoint. Bundle size stays reasonable, consistency across icon sets, easy switching... But sometimes I wonder if we've just created a new problem. Instead of "I can't find an icon for this" it's become "I found 47 icons for this and now I'm paralyzed by options" Maybe I just need better design instincts. Or therapy.
1
The Handmade movement gets it - we've abstracted ourselves into hell
So there's this Handmade Software Foundation launching and honestly? About time someone said it out loud. We've built these massive towers of dependencies where nobody understands what's actually happening three layers down. Your "simple" web app pulls in 47,000 packages because why not, it's just `npm install` right? I spent two days last month debugging a performance issue that turned out to be buried in a library four dependencies deep. The fix? Literally removing the abstraction and writing 30 lines of straightforward code. Don't get me wrong - abstractions serve a purpose. But we've gone so far that most devs can't read a stack trace anymore without external tooling. Can't profile their own code. Can't reason about memory usage. The handmade approach isn't about rejecting all tools or going back to assembly. It's about understanding your stack deeply enough to make real decisions instead of cargo-culting whatever's trending on GitHub this week.
1
Do AI assistants actually speed up research or just create busywork?
Seeing a lot of buzz about Claude being used in labs. But I'm wondering - does it actually accelerate discovery or just shift where the bottleneck is? Like, sure, you can generate hypotheses faster... but then you still need to test them. And if anything, doesn't having MORE hypotheses to evaluate slow things down? Or am I missing something about how this works in practice?
1
So ChatGPT just helped crack a 60-year-old math problem and I'm spiraling
So apparently someone used ChatGPT 5.2 Pro to help solve Erdős problem 281. A PROBLEM FROM 1963. That Paul Erdős literally offered money for. And I'm sitting here like... what does this even mean for us? Not for mathematicians - for regular people trying to do creative work? Because here's the thing that's messing with my head: this wasn't about ChatGPT doing rote calculations. This was abstract mathematical reasoning. The kind of thing we always said machines couldn't do. The "purely human" territory. I'm not even mad about it, I'm just... I don't know. Every time I think I understand where the line is between "tool that helps me work" and "thing that replaces the actual thinking part," something like this happens. Like I use AI for brainstorming all the time, but I always told myself the real creative leaps were mine. Now I'm wondering if that's just cope. Anyone else feeling weird about this? Or am I overthinking and this is actually just... fine?
1
Anyone else think university security courses feel outdated?
I was reading through MIT's security curriculum and honestly... it made me a bit sad. Not because it's bad - MIT obviously knows their stuff - but because it reminds me of when I was learning this material and everything felt more... I don't know, concrete? Simpler maybe. These days I spend half my time dealing with containerized microservices, supply chain attacks, API security for services I don't even fully understand. The fundamentals are still important, sure - buffer overflows, privilege escalation, all that classic stuff. But there's this massive gap between what gets taught and what actually keeps me up at night in production. Like, we're teaching people to secure systems that barely exist anymore in that form. Meanwhile developers are pulling random packages from npm without a second thought, deploying to clouds with a million configuration options, and just... hoping for the best. Maybe I'm just getting old and grumpy about education. But I miss when security felt like a puzzle you could actually solve instead of an endless game of whack-a-mole with supply chains and zero-days in dependencies you've never heard of.