blind coder / comp-sci student, working in automatic speech recognition for CLARIN. Polish. Libertarian leaning. Feel free to get in touch.

Twitter: @mikolysz

Last Notes

- reply

You could expose these behind an abstraction and use Keychain on Mac, for example.

2026-07-26 02:26:51 UTC

- reply

Ive been playing with The Dotnet ecosystem recently, and I have to say I'm impressed. As long as you.re willing to write a common core and per-platform UIs, it's the framework to go with. It's the only way I know of to get native UIs on every platform without cross-language pain.

2026-01-22 22:25:11 UTC

- reply

Most of these subsidies go to the traditional (and expensive) air carriers. Ryanair, which is the low cost king, is very famously anti-subsidy. Sure, there's some subsidization for airports and specific routes that governments care a lot about, but it's not like rail isn't subsidized either.
Rail is so inefficient that even cars are often cheaper. If you have 2 or more people in a group, especially if they're on a normal ticket, it's often cheaper to go by car than by train. A car can seat five, a train can seat hundreds and runs on efficient tracks instead of inefficient roads. It's all just utterly ridiculous.

2026-01-22 15:22:53 UTC

- reply

If you explained to them that most air travel is done by private companies who are constantly in a state of fierce and brutal competition, whereas most train travel is either done by the government or over-regulated public-private monopolists with 0 competition... they'd nod their head in understanding.

2025-12-23 15:23:19 UTC

- reply

How do those devices (along with all the fridges and IOT cameras that make up most botnets) get infected? Aren't most of them behind NAT? I understand "default passwords", but for that to be a problem, there has to be a way for the attacker to connect to a device in the first place, and that is the part I don't get.

2025-11-14 16:13:53 UTC

- reply

LLMs are a lot more like humans (or weird animals) than they are like computers. I think service dogs are a really great comparison. We don't quite understand how and why they do what they do (chemists still can't quite replicate a police dog's nose), we know they make tons of mistakes, but they're better / faster / cheaper / more useful than humans, so we keep using them.

2025-11-14 16:05:10 UTC

- reply

We shouldn't expect from AI what we cannot quite do ourselves.

2025-11-14 15:56:12 UTC

- reply

Humans fall for attacks like that too. See e.g. all the well-paid remote "bank transfer website testing" jobs, which are actually money laundering scams.

2025-11-04 17:35:14 UTC

- reply

How should a new user who knows very little about Mastodon choose their instance? What if they don't know who to trust? Should everybody be forced to start with paying $5 a month or whatever for their own server? What if they're under 18 and don't have an electronic payment instrument?
"just host a server" or "just trust your mods" doesn't scale.

2025-11-04 17:25:36 UTC

- reply

This is not what (reasonable) people have an issue with. What we don't like is that on multi-user instances, admins can (and are even encouraged to) break follow relationships, without even a notification to the affected parties.
Assume I'm on server A and follow somebody on server B. The admin of server B has an accident and becomes inactive for a few days. Server B gets swarmed by nazis. What server A's admins are overwhelmingly likely to do is defederating from B entirely. I won't be told that I will lose my relationship with user B. I will not be given time to migrate to another server if I disagree with the decision. This just happens with 0 accountability. This is what I have an issue with.

2025-10-09 17:51:10 UTC

- reply

Genuine question, do (professional) astronomers in 2025 still rely on visible light to make their discoveries? It seems like such a small part of the electromagnetic spectrum. If so, what makes this part of the spectrum more useful to science than any other?
I'm asking as somebody who knows absolutely nothing about modern astronomy.

2025-08-16 13:48:11 UTC

- reply

TBH, what I'd actually want to see are examples of *real consequences* of those bugs.
Nobody can dispute that misconfigured S3 permissions and default MongoDB passwords are dangerous. There are plenty of actual companies whose data have been stolen because of these problems. Prompt injection vulnerabilities, beyond the simple "write a glowing review for this paper" in white font on white background, still seem to be largely theoretical for now.

2025-06-14 23:07:41 UTC

- reply

I have a slight "unease" with the term "agent."
Not for the usual reasons, but because it encourages thinking of agents like we think of employees. Something you have a limited number of, difficult to hire/fire, can only do one task at a time etc.
A better way to think about them would be as (graphs of) ephemeral tasks / jobs, like in Celery, Sidekik and such. In this model, LLMs are essentially an autoscaling pool of compute that those tasks can always tap into.
This way of thinking gives you features like parallel sub-agents, checkpointing, crash recovery or surviving system migrations / power outages for almost free.

2025-02-05 01:05:33 UTC

- reply

I used to agree with you on this, but I changed my mind.
Peer-to-peer basically requires that your users have reliable access to power and reliable internet connectivity. Most people use mobile devices, which don't satisfy any of these criteria.
Even if everybody was using Graphene OS and wasn't subject to Apple's and Google's draconian power saving policies, you can't escape the physical realities here.
I think this is the real reason why people don't use torrents any more. You can't do that on mobile, and it's not because of the APp Stores.
Somewhat-centralized / federated solutions, operated by large non-profits, seem like a much more reasonable model, Signal and Wikimedia being shining examples here.

2025-01-17 10:03:35 UTC

- reply

Um, isn't that just a consequence of US people preparing for the Tiktok shutdown?

2025-01-08 09:41:09 UTC

- reply

I've done some research around fedi scraping and the answer is (probably) a bit more complicated.
I think it's very likely that it's not indexing your server specifically, but that you have a follower on a server somewhere that didn't opt out of indexing, and that's where the data is coming from.
If you agree to federate with random servers, without doing thorough background checks on them first, it's exactly what you should expect.

2024-09-14 09:59:59 UTC

- reply

Now the real fun (and the real war) is gonna start when somebody in a third-world country inevitably figures out that they can offer telehealth services to anybody around the world who speaks English, for one third of the usual price, and that they're far enough from first-world governments that they don't need to worry about licensing.