Malte Kliemann

Just translated my 𝐰𝐨𝐫𝐝 𝐬𝐞𝐠𝐦𝐞𝐧𝐭𝐚𝐭𝐢𝐨𝐧 𝐥𝐢𝐛𝐫𝐚𝐫𝐲 from C# to Rust. For English/ASCII text it worked right away, but for Chinese/Unicode, there are differences between C# and Rust in how the index and length of strings work. While in C# an index points to a char, in Rust, it points to a byte, and a Unicode char might consist of several bytes. In C# the length of a string is the number of chars, in Rust it is the number of bytes. E.g. "你好" has a length=2 in C#, but 6 in Rust. After those adjustments and changing the dictionary from English to simplified Chinese, it was able to do Chinese word segmentation correctly: from "部分居民生活水平" to ["部分","居民","生活",水平"] This will be integrated into SeekStorm to support indexing and searching of documents (JSON/PDF) in Chinese. https://lnkd.in/eYW9XgQk

7

I wrote a very simple web server (https://lnkd.in/dhKymSs3) containing just a single handler func registered against the DefaultServeMux, and compiled for the following Go environment variables GOOS=darwin GOARCH=amd64. The compiler spit out a 7MB binary. To provide a relative context, an equivalent program written in Rust (https://lnkd.in/d9f9cVnq) compiles down to a 2.9MB with the --release flag. I figured the go binary is relatively large due to static linking. This means that go runtime is bundled in and parts of Go standard library is bundled in. Plus, Go binaries include debugging information. All adding up to the relatively large executable file size. One potent step to reduce the compiled binary size, by at least 25% for a release build, is to strip off the debug information using the `go build` linker flags `-ldflags="-s -w"`. The debug information plays a vital role in helping to generate a human-readable stacktrace whenever a panic occurs. It also helps profiling tools (like pprof, trace, etc) to map performance data (like CPU and memory usage) back to the specific functions or lines of code that are responsible. Apparently, debug information is important during development, but might not be necessary in production. Omitting the DWARF debug information and stripping off the debug symbol table reduced the binary file to 4.9MB, a 30% reduction. We can further analyze the binary using `go tool nm` to see the what symbols are taking up the most space. On the bright side, everything the binary needs to run is self contained. No DLLs, no external dependencies. So unless it is imperative you produce a small binary, I'd say don't not sweat it.

6

“Public money, public code” ❓ Switzerland has made a groundbreaking move by requiring all government-funded software to be open source. This marks a huge step forward in the “public money, public code” approach. Not only does it promote transparency and security, but it also fosters innovation and collaboration within the public sector. It's a bold shift that encourages reduced vendor lock-in, boosts competition, and can lead to lower IT costs. Seeing public funds being used to create resources that are available to everyone aligns with the values of openness and accessibility that many of us in the tech community support. European countries are leading the charge, with Switzerland and Germany making significant strides, while the US remains more cautious. However, the potential here is massive – especially when it comes to the reuse of software and data in a way that benefits all. Open source isn’t just for hobbyists or small projects anymore – it’s becoming a vital part of how governments operate, and I’m excited to see what comes next. Governments worldwide could learn a lot from this initiative, and I believe we’re just scratching the surface of what’s possible when public resources are made available to the public. #OpenSource #GovernmentTech #Innovation #Transparency

9

7 Comments

𝗘𝘃𝗲𝗿 𝗯𝗲𝗲𝗻 𝘁𝗼𝗹𝗱 𝘁𝗵𝗮𝘁 𝘆𝗼𝘂𝗿 𝗽𝗮𝘀𝘀𝗶𝗼𝗻 𝗶𝘀 𝗮 𝗱𝗲𝗮𝗱 𝗲𝗻𝗱? 𝗛𝗮𝘃𝗲 𝘆𝗼𝘂 𝗲𝘃𝗲𝗿 𝗯𝗲𝗲𝗻 𝘁𝗼𝗹𝗱 𝘁𝗵𝗮𝘁 𝘆𝗼𝘂𝗿 𝗽𝗮𝘀𝘀𝗶𝗼𝗻 𝗶𝘀 𝗮 𝗱𝗲𝗮𝗱 𝗲𝗻𝗱? That's what happened to me 20 years ago when my book deal on C++ , after half of it was completed, fell through. The publisher's verdict: "C++ is dying, nobody uses it anymore". But here's the thing: I knew better. C++ wasn't dying, it was just taking a break to prepare for its glorious comeback a few years later as C++11. So I decided to defy the naysayers and forge my own path. 𝗙𝗮𝘀𝘁 𝗳𝗼𝗿𝘄𝗮𝗿𝗱 𝘁𝗼 𝘁𝗼𝗱𝗮𝘆: - I started the POCO C++ Libraries project in 2004 - proving the enduring value of C++, now used by thousands of developers worldwide.  - A few years later, I launched the macchina.io technology stack for IoT - based on the POCO C++ Libraries, now used in millions of devices around the world. - Now, I build AIoT products with my partners at 21strategies relying heavily on C++ for its performance and rock-solid stability, a key ingredient for success in this dynamic field. 𝗧𝗵𝗲 𝗺𝗼𝗿𝗮𝗹 𝗼𝗳 𝘁𝗵𝗲 𝘀𝘁𝗼𝗿𝘆? Don't let anyone tell you that your passions are outdated. Embrace the power of your own knowledge and keep doing what you do best! Curious about how C++ and AI are revolutionizing manufacturing? Check out my recently published use case that demonstrates the power of AI in manufacturing. #aiot #algorithms #EdgeComputing #TechTrends #StayTuned #emergingtechnologies #iotdevelopment #iiot

15

4 Comments

The PrEis Is Right! (i.e. public repo is [finally] up) (EisDoomLauncher app repo has been up for a few days) Almost all docu for build-&-run is IOU (as it's subject to change); not that I expect anyone besides me to be working with these codebases for awhile (if ever), but I plan to realize such docu for folks who want to play with it. The TL;DR version is: {ref preis deps, build preis, ref preis jar in a consuming app repo, build app, run its launcher `BAT`}. I began fully implementing what would become the 'PrEis Suite' the same day Hexagon Mining laid me of (due obviously to the sudden increase of time and decrease of existing {i.e. work} projects). Ergo it's ironically become a gesture of resilience, persistence, and perseverance; as well as a return to R&D ops under my brand, R&D process, and otherwise control. And it's not done. Another repo for another app (EiSpriteViewer) is pending in a day or two; as I need to first realize the 'JARBAT' solution therewithin. Then there are about a dozen 'TODO' actionable tasks amongst the 3 codebases; and twice that for the backlog and longer-term 'suspendlog'. There are also 4 other [GZ]Doom util app[let]s spec'd and scope'd. Thanks to those who've been liking prior update/progress posts; I truly do appreciate it! (i.e. hi Iago Alves da Silva 😎) https://lnkd.in/gxxbhQuw

3

2 Comments

🔥We live in golden times for developers! 🌟💻 Today I needed an image to be anonymized by removing facial features (since I wanted to try out https://lnkd.in/eeh2wN4r with a private image without actually publishing it). I asked ChatGPT 🤖 how to do it and immediately got a little Python program 🐍 (see below, example results shown in image which is not my private foto 😊) able to find all faces and blur them with a Gaussian filter. Face-finding has been a hard research problem three decades ago 🧠📚 and is still something a normal programmer cannot do just using plain programming. But now we have it (and many other powerful AI methods) at our fingertips via packages/libraries. 🛠️✨ Awesome, if you think about it! 🚀 (example computed on Google Colab using a T4 GPU) --------- anonymizer program below ----------- # Load the uploaded image image_path = "./4men.png" image = face_recognition.load_image_file(image_path) pil_image = Image.fromarray(image) # Find all face locations in the image face_locations = face_recognition.face_locations(image) # Create a copy of the image to draw the blurred regions draw = ImageDraw.Draw(pil_image) # Blur each face in the image for face_location in face_locations:     top, right, bottom, left = face_location     face_image = pil_image.crop((left, top, right, bottom))     # Use ImageFilter.GaussianBlur instead of Image.Filter.GaussianBlur     face_image = face_image.filter(ImageFilter.GaussianBlur(radius=15))       pil_image.paste(face_image, (left, top, right, bottom)) # Save and display the blurred image output_path = "./blurred_faces_image.jpg" pil_image.save(output_path) output_path

7

2 Comments

Hey, everyone on LinkedIn! How is it going? Today, I want to showcase Webserv, the most recent project I've completed at 42 Wolfsburg. Wait, but what is "Webserv"? Webserv is a project in the 42 Wolfsburg core curriculum focused on creating your own HTTP server. It is capable of initializing infinite (well, infinite as far as your system allows) servers as well as handling a great number of clients simultaneously. The technology behind it allows handling all the I/O operations asynchronously,  ensuring that no single request can block the entire server. Moreover, it is capable of handling static and dynamic content, allowing the creation of interactive and dynamic web applications. Anyway, there's a lot I can say about this project, but, as they say, "it's better to see it once than to hear about it a hundred times". Feel free to check out the link to the project on my Github and onward to the next accomplishment! #42Wolfsburg #Programming #CodingJourney

20

Don't Blame the Tool: A Lighthearted Look at the Recent Crowdstrike Bug Let's be honest, the recent Crowdstrike update was...well, engaging. Hold the applause for the blue screens everyone! While the pointer issue definitely caused some headaches, it's important not to throw C/C++ under the bus. Here's the thing: C/C++ are powerful languages, but demand responsibility. They offer fine-grained memory control, fantastic for performance, but also a double-edged sword if not handled carefully. The Crowdstrike bug seems like a classic case of a null pointer mishap, a coding error, not a language flaw. Modern C/C++ practices are your friend. There's a wealth of best practices and libraries out there to specifically prevent null pointer exceptions. Safer alternatives exist, and it seems Crowdstrike might not have taken full advantage of them in this instance. C/C++ powers the world (well, a lot of it). From operating systems to high-performance applications, C/C++ is the bedrock of countless critical systems. In the right hands, it's incredibly stable and reliable. The Crowdstrike incident feels more like a "how not to code in C/C++" lesson than a reason to ditch the languages entirely. It's like blaming a hammer for a smashed thumb – the issue lies in how it's used. #Crowdstrike #NullPointerParty #CppLove P.S. For those interested in bulletproofing their C/C++ code against null pointer woes, check out some of the fantastic resources available online!

2

1 Comment

What do Open Source Licences actually mean? I digged through it for the last couple of weeks and here is what I learned. Especially about GPLv3. We at CORS GmbH and CoreShop strongly believe in the Power of Open Source. But it’s also hard to build a Business Model around Open Source. Especially when you depend on other Open Source projects. Most developers don’t care about licensing and usually choose something like MIT or Apache. But in some cases, you are required to choose GPL or similar licenses like OSL or Mozilla. GPL means that any derivative work, needs to be the same license. Meaning if you fork a GPL product like Pimcore or Wordpress, it has to be GPL again. But what about plugins? And what defines a derivative work? That is the main mystery, but can be answered with this: is the plugin invoked by just calling an entry point? Then no, it is not a derivative work. But if the plugin is using classes, data-structure or functions from the main program, it is a derivative work cause it only works as a combined program. (Wordpress specifically states that plugins have to be GPL again) The GPLs main focus is about Open Source. Open Source does not mean to be open or free. But it means if you get the program, you have to get the source code as well and you can freely do what you want with it. Including of modifying, adapting, or even selling it again. But always under the terms of and as GPL. Why am I looking into this in the first place? Well, Pimcore is GPL and therefore CoreShop has to be GPL. How can you create a business model around GPL extensions that everyone can freely redistribute again? Free or paid. Dual-Licensing is the way to go. Like QT or MySQL do as well. Pimcore also chose this path. So you can get a commercial license from Pimcore (the PCL) and then you are not obliged to the GPL anymore. The PCL itself also forces you to not use Pimcore community edition (therefore the GPL version) anymore. Which also does not allow you to install Community extension under a PCL. (Except for when these are dual licensed too) Licenses are a weird and messed up topic for developers. Especially Copyleft Licenses. Fun fact: the OSL 3, which Magento is licensed with, states under 5. External Deployment that you have to distribute the core code with every installation that is online and publicly accessible? So every core change you make, has to be made available. 😱 What does this mean for us at CoreShop? Well, the same as for everybody else using GPL. Once we sell extensions, everybody can re-distribute them free or paid. So what did we do? We also dual licensed years ago and therefore we can sell extensions for projects that are PCL without the GPL issues anymore.

18

8 Comments

Good news for Russian bot farms. I wrote about prompt injections a couple of weeks ago here: https://lnkd.in/eZjQWfMQ) And here: https://lnkd.in/eZSgEuhZ. This is where someone submitting information to a model includes instructions to the model that override the default programming of the model. So a model tasked with summarising a text could be directed to output propaganda in support of the Morris Ring (the National Association of Morris Dancers). This is quite invidious since any user submitting content that is assessed by the model can hide instructions within the text or within images and the model will respond accordingly. OpenAI has been alive to this problem by using a hierarchy of instruction, effectively making sure that the model privileges the instructions set by the system, in this case OpenAI, ahead of a casual user of the system. The promise is that OpenAI will adopt this technique in its next release, effectively closing the most obvious prompt injection loopholes. It remains to be seen what vulnerabilities are left after this upgrade and how quickly researchers develop exploits that either overturn the primacy of the system prompt or fool the LLM into thinking they have the same status as a system prompt. In the meanwhile, if you’re managing a social media bot farm you can allow your cold, black heart to rest easy at night. Download our guide to using AI in bidding here: https://lnkd.in/eEafimtM We’re a start up that makes it easier to find business opportunities around the world. Our customers can access the $13tn public sector market and set up alerts so that they never miss an opportunity. Sign up for a 7-day free trial here: https://openopps.com/

15

Happy new year! Want fireworks in your Terminal? I got you 🔥 Try`npx firew0rks` in your command-line terminal. I published it today (thanks to some heavy inspirational from the Python ASCII art community). A few things you can do with it: Configurable loops: `npx firew0rks fireworks 30` Play fireworks infinitely: `npx firew0rks fireworks -1` Play a cosy fireplace: `npx firew0rks fireplace -1` Source: https://lnkd.in/gYKf6k2k #programming #softwareengineering #javascript

6,313

166 Comments

Look what's in C++23 Labels at the end of compound statements (P2324R2), e.g. void f(int& x) {    if (x)      goto END;   x = 42; END: }

20

41 Comments

Today, at OpenAI DevDay in San Francisco, big announcements were made. One of which is "Prompt Caching". This feature allows developers to avoid passing long prompts again and again. By reusing recently seen input tokens, developers can get a 50% discount and faster prompt processing times. How it works Caching is enabled automatically for prompts that are 1024 tokens or longer. When you make an API request, the following steps occur: Cache Lookup: The system checks if the initial portion (prefix) of your prompt is stored in the cache. Cache Hit: If a matching prefix is found, the system uses the cached result. This significantly decreases latency and reduces costs. Cache Miss: If no matching prefix is found, the system processes your full prompt. After processing, the prefix of your prompt is cached for future requests. Cached prefixes generally remain active for 5 to 10 minutes of inactivity. However, during off-peak periods, caches may persist for up to one hour.

8

Open source is eating the world. I wanted to find out what German R&D powerhouses contribute to the movement. Here is what I found by browsing around in Github:   I used a very simple methodology and just tried to find the repo with most stars for each company. As many repos are hosted by specific departments or labs, the list might not be completely accurate. Here are the results (more details in the slideshow): 1️⃣ Cariad/VW: 6 stars 2️⃣ Bosch: 380 stars 3️⃣ Mercedes-Benz: 255 stars 4️⃣ SAP: 2900 stars   🤔   This perspective oversimplifies things: Github stars are a vanity metric, there are many more ways to contribute to FOSS than owning repos and very successful projects might have already been moved to an independent organization such as Apache. But: I think the result is still devastating ‼️ My intention is not to point any fingers. I know from experience that it can be difficult to make the business economics work in the short term. I just think the issue is worth discussing and FOSS is a critical factor in today’s competitive landscape. There are so many advantages of open source, especially for larger organizations: 1️⃣ No vendor lock-in, especially with regards to cloud services 2️⃣ Better efficiency and security than custom solutions 3️⃣ Collaboration opportunities with startups and academia   So, please help me: Which open source success stories from large German corporates do you know (can be collaborations with startups / academia)? What do we have to do to get more of them?   Very much looking forward to learning what’s your perspective on this 👇in the comments ☕during a virtual coffee talk: https://lnkd.in/gjX6tzXJ #opensource #foss #industrialai

34

16 Comments

🚨 4:23 AM: The Battle with Gradle 🚨 Yes, it’s 4:23 AM, and I’m still here, battling through a stubborn "Gradle Error" that decided to crash my app build. Hours of effort—updating plugins, tweaking gradle-wrapper.properties—and finally, the app is back on track. Coding isn’t just about knowing how to fix things; it’s about persistence, patience, and understanding why issues happen. Tonight (or should I say this morning), I pushed through, learned a ton, and emerged a little stronger as a developer. Sometimes the late nights pay off in lessons learned. Here’s to the grind and the growth that comes with it. #code

45

3 Comments

[UA Insight] Main format of Creatives from Parrot Translator: 1. Real-person explanations combined with feature demonstrations 2. Humorous plot-based presentations 3. Real-life scenario feature demonstrations History repeats itself. So you might wanna pay attention.

1

I don't think Anthropic gets enough credit for their Claude 3.5 Sonnet model. 👏 Compared to ChatGPT 4o, which is often overly verbose and hallucinates on overly technical topics, Claude 3.5 Sonnet has impressed me with: * Concise responses: Unlike ChatGPT 4o that drowns you in unnecessary words, Claude gives much more concise responses. * Technical accuracy: I've found it remarkably reliable on technical matters, while avoiding the hallucinations. * Adaptability: Whether it's a quick question or an in-depth analysis, Claude adjusts its response style appropriately. I don't sign up for subscription services lightly, but Claude 3.5 Sonnet has been an exception for me. It's refreshing to interact with an AI that is both highly capable while concise. Have you tried Claude 3.5 Sonnet? What's been your impression so far? #AI #GenAI #MachineLearning #ChatGPT #LLM

10

4 Comments

Remember Perl? This is a great post about why #Perl is still relevant today! #Perl #ProgrammingLanguages #Tech #Automation #CPAN #Scripting #Developers #Coding

1