HNHacker News
TopNewBestAskShowJobs

arendtio

3,337 karma · joined June 8, 2018

submissionscomments
arendtio··on Show HN: KidScreen, a finite YouTube shelf chosen by parents
I like the idea.

I solved the issue by building an extra Android Launcher for the kids, where you can add apps and videos, but then you have the problem that you can't configure your own launcher on all tablets (e.g. Fire Tablets).

arendtio··on uBlock Origin Is Giving Up the Fight to Keep Ads Off Facebook
Actually, 10 years ago, I knew someone with roughly that job description (neither Google nor Facebook).
arendtio··on Docker Sandboxes – Disposable, isolated sandboxes for AI agents
> If an agent fires up NPM... :D :D

I think this is one of the big issues with today's MCP servers. Most are based on Node.js and take a lot more memory than they should, compared to the complexity that the job requires. Just run a few MCP servers locally, and all your RAM is gone...

arendtio··on Windows 11's built-in Weather app wastes more than 1 GB of RAM
I think we have to translate '1GB' here. It sounds like so little since it is just one GB, when in fact it is 8,000,000,000 bits.

When was the last time you counted to eight billion and thought it would be a good idea to build an app that would require that many bits of information to tell someone the weather forecast for the next week?

And this comparison is somewhat unfair since all numbers from 0 to 8 billion can be represented with just 33 bits...

arendtio··on New Mexico court orders Meta to pay $567m over harms to children’s mental health
But isn't a fine meant to hurt? I mean, isn't this the reason we have those X% of Y fines? So it seems less relevant how many people were harmed than who harmed them.
arendtio··on Danish high schoolers will have to verbally defend written assignments
I think the important part is having a good learning curve. Having to speak in front of 100+ people without any prior experience is terrifying.

However, there are very challenging tasks that can help to build experience in smaller groups. Like talking about a topic you just received a second earlier, without preparation, in a group of 10 where everybody takes turns and knows the experience from both sides.

arendtio··on Managing AI Coding Costs at Scale
I think the important aspects are making individual developers responsible for their costs and restricting the most expensive models when costs are too high. That way, people have a chance to learn when they tend to use the most expensive models for trivial tasks.
arendtio··on What happens if an entire class of workers loses faith in their careers
I think we shouldn't be depressed, but work on ideas and concepts of how this could be changed into something good.

I see a lot of people hyping AI and condemning it, but very few seem to be interested in working on what our future should look like.

arendtio··on What happens if an entire class of workers loses faith in their careers
I think it depends not just on culture but also on the volume of the conversation. Some people manage to have a conversation when you can barely hear them; others seem to have a conversation with everyone on the train.
arendtio··on Show HN: Skyline Co-op is like SimTower, but built around democratic socialism
If it was generated pre-2020 by some slow meatballs (like news.ycombinator.com), it is okay to have edges; post-2020, it must be closed to perfection.

I don't know what everybody is complaining about AI-generated content. In the past, nobody cared how the content was generated (ed, vim, emacs, VS Code, ...). It either was good or it was not.

Personally, I don't care if the content is AI-generated as long as it is good. But maybe that is, because I think that it is possible to generate good content using AI.

arendtio··on Position: LLMs Can't Jump
I don't think that the 'lossy' part is the problem. Instead, the problem is that language is only a subset of the human experience. So there are aspects that are not included when training models based on language. Making the models multi-modal helps to close that gap, but it still exists.

But ultimately this is just a matter of training data. I do not say that it is easy to obtain the required data, but it is not a fundamental problem LLMs can't overcome.

arendtio··on Android May Soon Restrict On-Device ADB
Every time I want to connect ADB to a device, it takes me far more time than it should. Having different ports for pairing and connecting is already a pain. I wonder what new hurdles will come up once they restrict adb to wlan0, even though atm I can't see any problems for my use cases (like pushing APKs, reading logs or connecting dev tools).

Not being able to use VPNs might be a problem in corporate setups (e.g. debugging an issue in production environments).

arendtio··on If coding has been solved, why does software keep getting worse?
My Arch Linux KDE Desktop with the Haswell i7-4770S CPU (from 2013) actually feels snappier than MacOS on my MacBook Pro M4. Sure, the MacBook has far superior performance overall, but if you just want to move some windows or get a response from a right-click, it just feels slower.
arendtio··on Be skeptical of OpenAI's rogue hacker agent story
This whole thing is just a PR stunt. I believe that they did not stage it intentionally, but to be honest, the hurdles they put in place were not that hard.

When I heard what the model did, I wasn't surprised one bit. When I read that it was 'unprecedented', I think yes, that is right, but not because it wasn't possible before, but because nobody else did it yet. I am sure that the Anthropic models could have done the same for a few months.

So probably just an experiment gone wrong, and the PR department found this an opportunity to capitalise on (literally).

arendtio··on What if we made advertising illegal?
I think forbidding it completely is somewhat radical, and while I am not a particular fan of advertisements, I think the real problem is personalized/targeted advertising.

Without personalized advertising, the surveillance of users would be much less lucrative. And if the same rules applied to all players, it should not have such a bad impact on the advertising market overall.

arendtio··on Self-host your mail server
I have had my own e-mail server since 2017, and there is only one instance where I know I had a delivery problem, and in that case it was funny, because I wrote to two people who had the same e-mail server, and one got the e-mail the other did not. So probably a SPAM issue, but certainly nothing my e-mail server could have done better...

AFAIK, that case in 2019 was the only time an e-mail went missing.

arendtio··on Buz – A fork of Bun using modern Zig, with sub-1s incremental builds
But isn't this a problem with the ecosystem and not with the program? I mean, with yarn / etc., you have the same problems, don't you?
arendtio··on Buz – A fork of Bun using modern Zig, with sub-1s incremental builds
Well, do one thing an one thing good as in Unix philosophy?

- Runtime

- Package manager

- Test runner

- Build tool

arendtio··on I Tried Building a Real App with AI. It Took a Year
I am not exactly sure, but my gut feeling is that Cursor existed before Claude Code was released, and the term vibe coding existed already at that time.

EDIT: Okay, Andrej tweeted about Vibe Coding on February 3rd 2025, and Claude Code was released in the same month. But Cursor was already there (I can see usage starting January 2025 in my account and know that it took me a few weeks before I tried it).

arendtio··on Buz – A fork of Bun using modern Zig, with sub-1s incremental builds
Why is there so much buzz around bun? Can't we just go back to node + npm + vitest + vite?
arendtio··on I think you might be fooling yourself with AI
I think it just takes time to integrate it into existing structures, which is a good thing in this case.
arendtio··on I think you might be fooling yourself with AI
> intentionally trying to mislead people

Is a bit hard. But I agree that the field is moving very fast and that it is hard to find studies that are up to date on the latest tooling.

arendtio··on I think you might be fooling yourself with AI
Maybe "using AI" is not sufficient by itself. Maybe you need to use it in the right way/setup/environment.
arendtio··on I think you might be fooling yourself with AI
I think the potential productivity gains are real, but they have a cumulative effect. Currently you can speed up the coding easily, but other tasks, like analysing logs in production for errors, still delay you. Once you add that capability, everything goes a little faster, but then you are being delayed by user acceptance testing; once you automate that, you are being delayed by requirements engineering, etc.

Every step is part of the journey of realising the productivity gains. If you want to, you can automate every step, and you won't just get a little faster. However, at the moment, the gains of being more productive when writing code are still eaten up by more time-consuming tasks like analysing bugs or having to solve organizational issues.

So it will take a while (and I think that is a good thing) until productivity gains are fully visible. For me, it is currently at work as if I had twice the amount of time I had in the past, because many of the repetitive tasks can be done much faster with AI. It is a bit like 60% of the tasks can be done in 10% of the time. But it already took me a few weeks to assemble MCP servers for many of the systems I use (in a responsible manner).

Even though I have a lot of fun building AI systems, I am very worried about the future of work, because I have the feeling that politics is busy with other drama instead of finding solutions for the time when corporations won't need human capital anymore.

arendtio··on Tell HN: Namecheap gave my account to an unverified third party
Honest citizens think like that; others wonder what the punishment is, even if it is illegal. I mean, maybe paying a fine vs maybe getting access to a popular domain?

At least corporations seem to work like that. Not following the law seems completely okay as long as the fine is not X% of their annual turnover.

arendtio··on Everyone should know SIMD
Everyone should know SIMD, okay, but please don't introduce that kind of code into my codebase for a linear 5x improvement.

If you work on a performance-critical application, that's fine, but most of the time we use some interpreted / VM / garbage-collected slow environment that performs so much slower than just using hardware directly, and we are fine with it. Under those circumstances, it is much better to have code that every developer can understand than to introduce a 5x improvement, with code that only the person who wrote it understands.

I mean, e.g. for Go, this simplicity was one of the explicit design goals. For C / Rust / Zig, it might be a different story, but I hope you choose those languages only if they fit your use-case.

arendtio··on Writing by hand is good for your brain
That is why I call good pens like my Fink [1] a thinking tool. I always wonder if it is nature or nurture, but I am sure that writing helps me to think in a straight line.

[1]: https://www.stefanfink.de/en/writing-instruments

arendtio··on Show HN: Ex Situ – Open-source spatial index of displaced cultural artifacts
> Coming from a design background

From a UX perspective, it is very weird to place the toggle button for the right panel at the top of the left panel... After I gave up to find a way to close the right panel, I found that button a few minutes later, when I looked around for other interactive elements.

arendtio··on Ente – Opening Our Books
Well, not E2EE, but I like Photoprism very much for the self-hosted photosharing use-case. In fact, I tried Ente a while ago and am always impressed by their communication skills, but from a product side, I prefer Photoprism.

I have a local instance running on a Raspberry Pi with about 149.000 photos inside.

arendtio··on What xAI's Grok build CLI sends to xAI: A wire-level analysis
AFAIK, Cursor does some kind of indexing locally. So they don't have to upload all files, but can still search through them in order to find the relevant _parts_ that they upload so that the model can use them.
← PreviousPage 2 of 34Next →