raphman

Just to be sure: your blog post mentions 24.12.2024 - did you mean 2025?

Is this the right connection?:

https://bahn.expert/details/NX%2028521/j/20251224-a0049123-9...

The RE5 seems to be operated by National Express, not Deutsche Bahn, right? (but DB InfraGo is most probably responsible for the routing)

Thanks! First hit online: https://www.lab404.com/3741/readings/sontag.pdf

Out of curiosity: what led you to write "Susan Sunday" instead of "Susan Sontag"? (for other readers: "Sonntag" is German for "Sunday")

Claude Code user¹ says Claude Code wrote continuously incorrect code for the last hour.

I asked it to write Python code to retrieve a list of Kanbord boards using the official API. I gave it a link to the API docs. First, it wrote a wrong JSONRPC call. Then it invented a Python API call that does not exist. In a new try, I I mentioned that there is an official Python package that it could use (which is prominently described in the API docs). Claude proceeded to search the web and then used the wrong API call. Only after prompting it again, it used the correct API call - but still used an inelegant approach.

I still find some value in using Claude Code but I'm much happier writing code myself and rather teach kids and colleagues how to do stuff correctly than a machine.

¹) me

FWIW, many of the researchers on the paper did not study in the U.S. but immigrated after their PhD studies.

I checked the first, middle, and last author: Lars Mescheder got his PhD in Germany, Bruno Lecouat got his PhD in France, Vladlen Koltun got his PhD in Israel.

(Edit: or maybe they did not actually immigrate but work remote and/or in Europe)

I found this quote from the paper by Catchpole et al. quite important:

"Whilst it is encouraging that analogies from other industries with longstanding cultures of safety and reliability may be extrapolated, it is also important to recognize the unique demands of health care. One crucial difference that emerged was that Formula 1 and aviation both have a relatively stable workforce, with minimal staff turnover. For example, out of about 20 members of a pit team, only one or two members change annually. In contrast, turnover of staff in health care is far higher, with six residents rotating every 3 months in the study unit, and a nurse turnover of approximately 10%."

Some more sources I saved recently after learning about Elliott's work:

The original paper by Catchpole et al. (2006): https://onlinelibrary.wiley.com/doi/10.1111/j.1460-9592.2006...

A quite interesting lecture on the whole project, given by Elliott in a racecar driver's suit: https://www.youtube.com/watch?v=ZeMGrHcfRjY

Annotated transcript of the talk available here: https://www.gresham.ac.uk/watch-now/formula-1-and-its-contri...

A WSJ article from 2006: A Hospital Races To Learn Lessons Of Ferrari Pit Stop: http://www.wsj.com/articles/SB116346916169622261 (archived copy: https://archive.is/xEbio)

Times article from 2024: The surgeon who used F1 pitstop techniques to save lives of babies: https://www.thetimes.com/sport/formula-one/article/professor... (archived copy: https://archive.is/1iGXK)

There are companies that laser-'etch' 3D images into glass. I guess it's not that hard to find one that accepts a list of xyz coordinates.

FWIW, the Haribo charger seems to be rather well made: https://www.youtube.com/watch?v=xT_t5nvFcoY (26:50, by ZeroBrain, a German YouTuber that disassembles electronic devices)

hiq's comment was not about Whisperfish but about the presage library. My comment can be read as "Whisperfish wrote their own implementation of the signal protocol" - which is wrong. (Sorry, can't edit it anymore)

With presage, Whisperfish has a high-level Rust library that wraps the official libsignal libraries (which are also written in Rust) and makes it much easier to write clients. The official Signal repo only contains Java/Typescript/Swift wrappers. As presage is rather obscure, I thought that some HN readers might appreciate the link.

The Whisperfish [1] project (a Signal messenger for Sailfish OS) maintains an independent Signal client library written in Rust [2]. It works quite well - unless Signal decides to change their protocols or kick non-standard clients.

[1] https://gitlab.com/whisperfish/whisperfish

[2] https://github.com/whisperfish/presage

My personal opinion (based in part on HN):

- In a nutshell: everyone should know about UML and how some relevant diagram types look. Nobody should use it religiously.

- Diagrams are great tools for communicating ideas and implementations on an abstract level, and for interactively thinking about ideas that can be drawn in some way. A diagram can also visually highlight problem areas (e.g., components with many dependencies).

- Good use cases for diagrams in SE are e.g., explaining high-level architecture to someone, thinking about and refining database schemas, documenting interaction flows, or giving an overview of class structure in automatically generated code documentation

- It is nice to have visual consistency - it makes parsing a diagram so much faster and less error-prone. UML provides sensible guidelines or inspiration on how to draw e.g., a sequence diagram or a class diagram. There is no need to re-invent the wheel.

- In most cases, knowing about the diagram types and their approximate design is enough. As probably very few people know what the different styles of arrows or boxes might mean, it is a good idea to annotate important elements with their meaning.

- For thinking/communication, drawing diagrams by hand is the best way. Excalidraw/tldraw are nice for stuff that should go into static documents.

- Documenting the state of a system with manually generated diagrams is tedious and requires constant updates. A better way is to auto-generate e.g., class diagrams using PlantUML or Mermaid.

- Nobody uses UML for designing complex systems and then auto-generating code.

Thanks HN! I regularly open HN during lectures. There is no better way to show my students what software engineering entails and why I focus on certain topics.

Is SCRUM really as great as its evangelists claim? Let's read HN comments.

What are good use cases for UML? Let's check out HN.

Does anyone actually care about CoCoMo or CMMI? Let's read ... oh - nearly nobody's talking about it there. Maybe it won't be that relevant to the students.

Dito. I 'grew up' with Bash in the early 2000s and never knew about the globstar. I an pretty sure that I never encountered it in any scripts or tutorials - otherwise I'd certainly have looked it up.

"SAT/SMT by Example" also contains many Z3 examples (and has a new URL): https://smt.st/

Given the code has been completely vibe-coded, what does this mean in practice?:

> Copyright (c) 2025

Whose copyright? IIRC, it is consensus that AI cannot create copyrightable works. If the author does not own the copyright, can they add a legally binding license? If not, does this have any legal meaning?:

> IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY

> "most idiotic devices anyone could own"

Ever been to Chesterton's Fence?

Hypothetically, some people who own such an idiotic device might have pets that bring in lots of dirt from the fields, lose lots of hair, and get a little bit agitated by the normal vacuum cleaner but more or less ignore the robot vacuum.

I can confirm this. We have been using it for 10 years now in our research lab. No data loss so far. Performance is great. Integration with OnlyOffice works quite well (there were sync problems a few years ago - I think upgrading OnlyOffice solved this issue).

FWIW, Seeed Studios offers a range of Intel SBCs with an embedded microcontroller that is explicitly meant to be used like an Arduino board [1].

[1] https://www.seeedstudio.com/ODYSSEY-X86-v2-board-p-5075.html

FWIW, I actually received my Jolla Tablet (albeit delayed). Your are right that Jolla is sometimes less transparent and professional than one would expect. However, I realize that Jolla is also under much more scrutiny by the community than other tech companies, and some people demand ridiculous levels of transparency/quality/features.

I have been using SailfishOS phones as my main driver for ten years now. Some random, personal, possibly uninformed thoughts:

- It is not for everyone. Some Linux experience and willingness to tinker with it is helpful.

- Despite the many limitations, I love the UI, the spirit, the platform, and the community. I fear the day where I have to switch to a different OS.

- Many Android apps can be run via the AlienDalvik/AppSupport middleware. However, raw BLE is not supported. Thus, most e-scooter apps won't work. My banking app runs okay-ish.

- Google Play Store and Google Play Services can be installed by following non-trivial tutorials. I don't use them.

- The hardware abstraction layer that makes proprietary Android drivers work with SailfishOS is cool.

- QML and C++/Python/JS allow for easy, rapid app development. The custom widgets have a unique, consistent, simple style.

- As most of the UI is written in QML, it is possible to adjust and extend most of the UI shell and the base applications just by editing these resource files on the phone. For example, one can add additional widgets to the lock screen or change animation speeds.

- A nice tool, Patch Manager allows transparently and reversibly applying such modifications. This is so cool, even though the patches often have to be adapted for each major OS version.

- Jolla, the Finnish company behind SailfishOS is tiny and had to let go a lot of engineers and supporting staff a few years ago. Development has slowed down significantly.

- There are about two dozen very active developers in the community who write awesome apps. There are native clients for Discord (no voice/video), Signal, Telegram, Slack, Mastodon, Hacker News, etc.

- Unfortunately, the browser is stuck with outdated Gecko (despite heroic efforts by a developer who upgraded it from ESR 78 to ESR 91 [1]).

- Only a few smartphones are supported by SailfishOS - either officially supported by Jolla (e.g., some Sony phones and some Jolla-branded ones) or supported via community ports. Often the hardware support is a little bit buggy.

EDIT: of course, if you visit the forums, you will see quite some criticism of Jolla - and some of it is well deserved. It would be great if there were better hardware, fewer bugs, better support for Android apps, etc. Personally, I feel that Jolla is really trying to make SailfishOS better but that they lack really stable sources of income and have made some less-than-ideal decisions in hindsight. The best solution would be to get EU funding for stabilizing the platform and finding a business model that generates recurring income from large organizations. Selling to private customers without being able to extract recurring income and being dependent on badly-documented hardware is not going to work.

[1] https://www.flypig.co.uk/?to=gecko&list_id=975&list=gecko

Thanks for pointing this out. If someone reports a bug in my software, I usually acknowledge the reporter in the commit message. Seems like vibe coding also allows one to get rid of this pesky obligation.

Honest question: why would this code clamp the reported round-trip time? By default, min = 0.05 ms and max = 800 ms [1].

           if (rtt < config.min_rtt)
                rtt = config.min_rtt;
            else if (rtt > config.max_rtt)
                rtt = config.max_rtt;

Wouldn't this hide bugs in the code or network anomalies? Replies from localhost seem to typically arrive in less than 50 µs.

Comments in an earlier version [2] make no sense to me:

            /* Use standard timersub for more accurate results */
                    if (rtt < 0)
                        rtt = 0;

                    /* Cap at reasonable maximum to handle outliers */
                    if (rtt > 1000)
                        rtt = 1000;

[1] https://github.com/davidesantangelo/fastrace/blob/5b843a197b...

[2] https://github.com/davidesantangelo/fastrace/commit/79d92744...

The commit messages (with dozens of semi- and unrelated changes in each commit) suggest so.

https://github.com/davidesantangelo/fastrace/commit/79d92744...

(For me, this does not necessarily say anything about code quality. However, if a whole project is AI-generated, the author has no enforceable copyright IMHO, and thus, the 2-clause BSD license is void.)

There is/was https://github.com/selfspy/selfspy (not updated in 10 years).

I'm doing this in the software engineering¹ course I teach. However:

a) Practical CI/CD requires understanding (and some practical experience) of many other concepts like Linux shell scripting, version control, build automation, containers, server administration, etc. As few students (in our degree programme) have sufficient experience in these concepts, I spend about half of the semester teaching such devops basics instead of actual software engineering.

b) Effectively, teaching CI/CD means teaching how GitHub or GitLab do CI/CD. I feel a little bit uncomfortable teaching people how to use tech stacks owned by a single company.

¹) Actually it's more a course on basic software craftsmanship for media informatics students because no such course exists in our curriculum and I find it more important that students learn this than that they understand the V model.

> OpenDoc

For anyone interested in the Apple future that could have been, check out Jim Miller's articles, e.g. on LiveDoc (https://www.miramontes.com/writing/livedoc/index.php)

No, the article incorrectly cites/summarizes the original blog post [1] which indeed seems to pull some numbers out of thin air:

> "I’m going to use a bunch of numbers here that I believe to be directionally correct,"

> "Let’s start with total datacenter spend for 2025. Insiders think it’s going to clock in at around $400 billion."

(in the US? worldwide?)

> "I am sure I’m off by a few percent in these categories, but I’m relying on AI and we all know it’s still imperfect."

> "the AI datacenters to be built in 2025 will suffer $40 billion of annual depreciation, while generating somewhere between $15 and $20 billion of revenue."

In a follow-up article [2], the author says that he learned that the investments probably have much less than the 10 years of depreciation time he originally assumed. Again, no external sources are linked.

> "Personal self-doubts disappeared, and high-placed individuals reached out to share their epiphanies "

So, it seems the author estimated how much money will be invested in AI datacenters in 2025. The 20 billion of annual revenue mentioned in the article are just the share of all AI revenue for these new data centers. And these numbers have just been estimated by the author - apparently with the help of AI.

While the article and the blog posts may have some merit, it seems all based on some guy's assumptions and conversations he claims to have had.

[1] https://pracap.com/global-crossing-reborn/

[2] https://pracap.com/an-ai-addendum/

> but Airline and big hotel websites have way too many dark patterns made to confuse the user and force them to pay extra.

And sooner or later these websites will implement new dark patterns to confuse the LLMs...

"Sora Update #4: Through a partnership with Google, Meta and Snap Inc., you will be able to generate tasteful photos of the cute girl you saw on the bus. She will receive a compensation of $0.007 once she signs our universal content creators' agreement."

Thanks - seems to work quite well.