The most exciting search engine 68k can handle.

There’s Nothing Boring About Web Search On Retro Amigas

October 31, 2025 by Tyler August 23 Comments

Do you have a classic Amiga computer? Do you want to search the web with iBrowse, but keep running into all that pesky modern HTML5 and HTTPS? In that case, [Nihirash] created BoingSearch.com just for you!

BoingSearch was explicitly inspired by [ActionRetro]’s FrogFind search portal, and works similarly in practice. From an end-user perspective, they’re quite similar: both serve as search engines and strip down the websites listed by the search to pure HTML so old browsers can handle it.

Boing search in its natural habitat, iBrowse on Amiga.

The biggest difference we can see betwixt the two is that FrogFind will link to images while BoingSearch either loads them inline or strips them out entirely, depending on the browser you test with and how the page was formatted to begin with. (Ironically, modern Firefox doesn’t get images from BoingSearch’s page simplifier.) BoingSearch also gives you the option of searching with DuckDuckGo or Google via the SerpAPI, though note that poor [Nihirash] is paying out-of-pocket for google searches.

BoingSearch is explicitly aimed at the iBrowse browser for late-stage Amigas, but should work equally well with any modern browser. Apparently this project only exists because FrogFind went down for a week, and without the distraction of retrocomptuer websurfing, [Nihirash] was able to bash out his own version from scratch in Rust. If you want to self-host or see how they did it, [Nihirash] put the code on GitHub under a donationware license.

If you’re scratching your head why on earth people are still going on about Amiga in 2025, here’s one take on it.

Why You Shouldn’t Trade Walter Cronkite For An LLM

October 30, 2025 by Tyler August 45 Comments

Has anyone noticed that news stories have gotten shorter and pithier over the past few decades, sometimes seeming like summaries of what you used to peruse? In spite of that, huge numbers of people are relying on large language model (LLM) “AI” tools to get their news in the form of summaries. According to a study by the BBC and European Broadcasting Union, 47% of people find news summaries helpful. Over a third of Britons say they trust LLM summaries, and they probably ought not to, according to the beeb and co.

It’s a problem we’ve discussed before: as OpenAI researchers themselves admit, hallucinations are unavoidable. This more recent BBC-led study took a microscope to LLM summaries in particular, to find out how often and how badly they were tainted by hallucination.

Not all of those errors were considered a big deal, but in 20% of cases (on average) there were “major issues”–though that’s more-or-less independent of which model was being used. If there’s good news here, it’s that those numbers are better than they were when the beeb last performed this exercise earlier in the year. The whole report is worth reading if you’re a toaster-lover interested in the state of the art. (Especially if you want to see if this human-produced summary works better than an LLM-derived one.) If you’re a luddite, by contrast, you can rest easy that your instincts not to trust clanks remains reasonable… for now.

Either way, for the moment, it might be best to restrict the LLM to game dialog, and leave the news to totally-trustworthy humans who never err.

Hello World In C Without Linking In Libraries

October 29, 2025 by Maya Posch 27 Comments

If there’s one constant with software developers, it is that sometimes they get bored. At these times, they tend to think dangerous thoughts, usually starting with ‘What if…’. Next you know, they have gone down a dark and winding rabbit hole and found themselves staring at something so amazing that the only natural conclusion that comes to mind is that while educational, it serves no immediate purpose.

The idea of applying this to snipping out the <stdio.h> header in C and the printf() function that it provides definitely is a good example here. Starting from the typical Hello World example in C, [Old Man Yells at Code] over at YouTube first takes us from the standard dynamically linked binary at a bloated 16 kB, to the statically linked version at an eyepopping 767 kB.

To remove any such dynamic linkages, and to keep file sizes somewhat sane, he then proceeds to first use the write()function from the <unistd.h> header, which does indeed cut out the <stdio.h> include, before doing the reasonable thing and removing all includes by rewriting the code in x86 assembly.

While this gets the final binary size down to 9 kB and needs no libraries to link with, it still performs a syscall, after setting appropriate register values, to hand control back to the kernel for doing the actual printing. If you try doing something similar with syscall(), you have to link in libc, so it might very well be that this is the real way to do Hello World without includes or linking in libraries. Plus the asm keyword is part of C, although one could argue that at this point you could just as well write everything in x86 ASM.

Of course, one cannot argue that this experience isn’t incredibly educational, and decidedly answers the original ‘What if…’ question.

Continue reading “Hello World In C Without Linking In Libraries” →

Making Math Less Stressful With A Python Super-Calculator

October 26, 2025 by John Elliot V 15 Comments

In a recent write-up, [David Delony] explains how he built a Wolfram Mathematica-like engine with Python.

Core to the system is SymPy for symbolic math support. [David] said being able to work with symbolic math easily has helped his understanding of calculus and linear algebra. For statistics support he includes NumPy, pandas, and SciPy. NumPy is useful for creating multidimensional arrays and supports basic descriptive statistics such as mean, median, and standard deviation; pandas is a library for operating on tabular data arranged into “DataFrames”, it can load data from spreadsheets (including Excel) and relational databases; and SciPy is a “grab bag” of operations designed for scientific computing, it includes some useful statistics operations, including common probability distributions, such as the binomial, normal, and Student’s t-distribution.

For regression analysis [David] includes statsmodels and Pingouin. If you’re not familiar with the term “regression analysis” it basically refers to the process of curve fitting. When your data is two-dimensional, with one dependent variable, the simple linear regression algorithm will generate a function that fits the data as y = mx + b, including the slope (m) and the y-intercept (b); this can be extrapolated to higher dimensional data and other types of regression.

If you have an interest in symbolic math you might enjoy learning about Mathematica And Wolfram On The Raspberry Pi.

What If Tinkercad Was Self-Hosted?

October 25, 2025 by Al Williams 31 Comments

While we use a lot of CAD tools, many of us are fans of Tinkercad — especially for working with kids or just doing something quick. But many people dislike having to work across the Internet with their work stored on someone’s servers. We get it. So does [CommonWealthRobotics], which offers CaDoodle. It is nearly a total clone of Tinkercad but runs on Windows, Linux, Mac, or even Chrome OS.

Is it exactly Tinkercad? No, but that’s not always a bad thing. For example, CaDoodle can work with Blender, FreeCAD, OpenSCAD, and more. However, on the business end, it sure looks like the core functions of Tinkercad.

The program appears fairly new, so you have to make some allowances. For example, the Linux AppImage seems to have difficulty loading plugins (which it needs to import many of its file formats). In addition, on at least some systems, you have to resize the window after it starts, or it won’t respond. But, overall, it is pretty impressive. The Settings, by the way, has a checkbox for advanced features, and there are some other goodies there, too.

One reason we found this interesting is that we sometimes go into schools, and they don’t want us to have kids on the Internet. Of course, they don’t like us installing random software either, so you can pick your battles.

Tinkercad, of course, continues to add features. Not all of which you’d expect in a drawing package.

Reverse Engineering STL Files With FreeCAD

October 21, 2025 by Al Williams 30 Comments

If you think about it, STL files are like PDF files. You usually create them using some other program, export them, and then expect them to print. But you rarely do serious editing on a PDF or an STL. But what if you don’t have anything but the STL? [The Savvy Engineer] has a method to help you if you need to reverse engineer an STL file in FreeCAD. Check it out in the video below.

The problem is, of course, that STLs are made up of numerous little triangles. The trick is to switch workbenches and create a shape from mesh. That gets you part of the way.

Once you have a shape, you can convert it to a solid. At that point, you can create a refined copy. This gives you a proper CAD file that you can export to a STEP file. From there, you can use it in FreeCAD or nearly any other CAD package you like to use.

Once you have a proper object, you can easily use it like any other solid body in your CAD program. This is one of those things you won’t need every day, but when you do need it, it’ll come in handy.

Want to up your FreeCAD game? We can help. There are other ways to hack up STL files. You can even import them into TinkerCAD to do simple things, but they still aren’t proper objects.

Continue reading “Reverse Engineering STL Files With FreeCAD” →

Nanochat Lets You Build Your Own Hackable LLM

October 20, 2025 by Donald Papp 23 Comments

Few people know LLMs (Large Language Models) as thoroughly as [Andrej Karpathy], and luckily for us all he expresses that in useful open-source projects. His latest is nanochat, which he bills as a way to create “the best ChatGPT $100 can buy”.

What is it, exactly? nanochat in a minimal and hackable software project — encapsulated in a single speedrun.sh script — for creating a simple ChatGPT clone from scratch, including web interface. The codebase is about 8,000 lines of clean, readable code with minimal dependencies, making every single part of the process accessible to be tampered with.

An accessible, end-to-end codebase for creating a simple ChatGPT clone makes every part of the process hackable.

The $100 is the cost of doing the computational grunt work of creating the model, which takes about 4 hours on a single NVIDIA 8XH100 GPU node. The result is a 1.9 billion parameter micro-model, trained on some 38 billion tokens from an open dataset. This model is, as [Andrej] describes in his announcement on X, a “little ChatGPT clone you can sort of talk to, and which can write stories/poems, answer simple questions.” A walk-through of what that whole process looks like makes it as easy as possible to get started.

Unsurprisingly, a mere $100 doesn’t create a meaningful competitor to modern commercial offerings. However, significant improvements can be had by scaling up the process. A $1,000 version (detailed here) is far more coherent and capable; able to solve simple math or coding problems and take multiple-choice tests.

[Andrej Karpathy]’s work lends itself well to modification and experimentation, and we’re sure this tool will be no exception. His past work includes a method of training a GPT-2 LLM using only pure C code, and years ago we saw his work on a character-based Recurrent Neural Network (mis)used to generate baroque music by cleverly representing MIDI events as text.