Transcribing The Source Of The First DOS For The IBM PC

April 30, 2026 by Maya Posch 9 Comments

Doing software archaeology can be a harrowing task, as rarely do you find complete snapshots of particular versions of software. Case in point the development of MS-DOS – also known as IBM PC DOS – from 86-DOS, which recently got a lucky break in the form of printed source listings. These printouts come courtesy of [Tim Paterson], the creator of 86-DOS and of MS-DOS during his time working for Microsoft.

These code listings contain the sources of the 86-DOS 1.00 kernel, multiple development snapshots, and also listings for utilities like CHKDSK. These printed listings additionally contain many handwritten notes, making transcribing it into working source code somewhat of a chore. The results can be found on the GitHub project page, with the original scans available on Archive.org.

Of the ten bundles of continuous feed paper prints all but two have been transcribed so far, though with the various DOS kernels and the Seattle Computer Products (SCP) assembler source already ready for compilation. This includes 86-DOS 1.00, MS-DOS 1.25 and PC-DOS 1.00-dev, requiring the same SCP assembler to create a binary.

In the project page README a number of blog posts are also linked that add even more technical detail. Anyone who wants to pitch in with transcribing and/or testing recovered source code is welcome to do so.

Why Model Collapse In LLMs Is Inevitable With Self-Learning

April 29, 2026 by Maya Posch 76 Comments

There is a persistent belief in the ‘AI’ community that large language models (LLMs) have the ability to learn and self-improve by tweaking the weights in their vector space. Although there’s scant evidence that tweaking a probability vector space is anything like the learning process in biological brains, we nevertheless get sold the idea that artificial general intelligence (AGI) is just around the corner if we do just enough tweaking.

Instead of emerging super intelligence, the most likely outcome is what is called model collapse, with a recent paper by [Hector Zenil] going over the details on why self-training/learning in LLMs and similar systems is a fool’s errand. For those who just want the brief summary with all the memes, [Metin] wrote a blog post covering the basics.

In the end an LLM as well as a diffusion model (DM) is a statistical model of input data using which a statistically likely output can be generated (inferred) based on an input query. It follows intuitively that by using said output to adjust the model with, the model will over time converge on a kind of statistical singularity rather than some ‘AI singularity’ event. This is also why these models need to be constantly trained with external, human-generated data in order to prevent such a collapse.

In the paper by [Hector] a mathematical model is created to demonstrate that an LLM, DM or similar statistical model undergoes degenerative dynamics whenever said external input is reduced. Although in the paper a mechanism is suggested to counter the entropy decay within the model, the ultimate point is that a statistical model cannot improve itself without continuous external anchoring.

The idea of LLMs being at all intelligent in any sense has been a contentious one, with the concept of language models being equated with ‘AI’ dating back to the 20th century, including as fun home computer projects. Much of the problem probably lies in humans projecting intelligent behavior onto these statistical models, turning LLMs into ‘counterfeit humans’, not helped by how closely generated text can resemble something written by a human, even if completely confabulated.

Thanks to [deshipu] for the tip.

How To Kill Humidity Sensors With Humidity

April 29, 2026 by Maya Posch 15 Comments

An often overlooked section in the datasheets for popular humidity sensors like the BME280 and DHT22 is the ‘non-condensing humidity’ bit, which puts an important constraint on which environments you can use this sensor in. This was the painful lesson that [Mellow Labs] recently had to learn when multiple of such sensors had kicked the bucket after being used in a nicely steamed-up bathroom. Fortunately, it introduced him to sensors that are rated for use in condensing humidity environments, such as the SHT40 that’s demonstrated in the video.

This particular sensor is made by Sensirion, and as we can see in the datasheet it features a built-in heater that allows it to keep working even in a condensing environment. This heater has three heating levels which are controlled via the I2C interface, though duration is limited to one second in order to prevent overheating the sensor.

Of note is that you cannot take measurements while the heater is operating, and its use obviously increases power draw significantly. This then mostly leaves when to turn on the heater as an exercise to the engineer, with [Mellow Labs] opting to start the heater when relative humidity hit 70% as a conservative choice.

In the comments to the video other options for suitable sensors were pitched, including the Bosch BME690 which is similarly rated for condensing environments. All of which condenses down to the importance of reading the datasheet for any part that you intend to use in possibly demanding environments.

Continue reading “How To Kill Humidity Sensors With Humidity” →

Recycling PLA And Other Plastic Waste With Compression Molding

April 28, 2026 by Maya Posch 4 Comments

After previously trying out low-tech compression molding with a toaster oven and 3D printed molds, [future things] is back with a video that seeks to explore some of the questions raised after the first video. Questions such as how well this method works with HDPE and PLA thermoplastics, whether the flashing could be cut off by the mold and the right temperatures and times to heat the plastic before a charge is ready for inserting into the mold.

In this video the same PHA-based mold is used, but in a three-piece configuration to allow for a more complex shape. This way game tokens could be made for use by the son of the author, which also shows one straightforward and very practical use of this method.

A big change here is that no more metal chopsticks are used to handle the charge, as this was found to cool down the heated plastic too much. Instead the hot charge is handled with fingers and wooden chopsticks, with the plastic heated until it has about the consistency of thick honey. For LDPE this takes about 5-7 minutes at 130°C. After compressing the charge into the mold, about 30 seconds are all it takes for the plastic to cool down enough.

There was a question about the use of mold release spray, but this didn’t seem to cause any issues, so can probably be used safely. As for other plastic types, HDPE works fine too when you heat it up at a slightly higher temperature and don’t mind it being tougher to handle.

Easiest is probably PLA, which would seem unsurprising. Using some chopped-up PLA printing waste it was easy enough to make a few more game tokens, demonstrating that this method is very viable for converting scrap FDM print waste into such items. As noted in the comments by [edmundchao] this method works great too for PETG, using PETG molds, while using a ratcheting clamp for extra pressure instead of just pressing by hand.

Continue reading “Recycling PLA And Other Plastic Waste With Compression Molding” →

Can You 3D Print A Pinball Machine That’s Fun To Play?

April 28, 2026 by Maya Posch 8 Comments

It seems fair to say that pinball machines are among the most universally loved gaming systems known today, yet the full-sized ones are both very expensive and very large, while even the good quality table-sized ones tend to be on the expensive side. That raises the question of whether a fully 3D printed pinball machine could at all be fun and not just feel like a cheapo toy? A recent video by [Steven] from [3D Printer Academy] on YouTube makes here a compelling argument that it might actually be worth something to consider.

In addition to being fully modular and customizable the most compelling element is probably that the design supports two- and four-player multiplayer. This sees the metal balls leaving at the rear and from there entering the playing field of another player’s machine, which can probably get pretty chaotic.

Unfortunately this is part of a Kickstarter campaign, so you’ll have to either shell out some cash to get access to the print files or DIY your own version. We’d also be remiss to not address the durability concerns of a 100% plastic pinball machine like this, plus the lack of serious heft to compensate for more enthusiastic playing styles.

If you are more into traditional DIY pinball machines, we have covered these as well, along with small screen-based machines, and their miniature brethren for when space is really at a premium.

Continue reading “Can You 3D Print A Pinball Machine That’s Fun To Play?” →

The Challenges Of 3D Printing Reliable Springs

April 27, 2026 by Maya Posch 19 Comments

Springs are great, but making them out of plastic tends to come with some downsides, for fairly obvious reasons. Creating a compliant mechanism that can be 3D printed and yet which doesn’t permanently deform or wear out after a few uses is therefore a bit of a struggle. The compliant toggle mechanism that [neotoy] designed is said to have addressed those issues, with the model available on Printables for anyone to give a shake.

The model in question is a toggle, which is the commonly seen plastic or metal device that clamps down on e.g. rope or cord and requires you to push on it to have it release said clamping force. Normally these use a metal spring inside, but this version is fully 3D printable and thus forms a practical way to test this particular compliant mechanism with a variety of materials.

The internal spring is a printed spiral spring, with the example in the video printed in PETG. You can of course also print it in other materials for different durability and springiness properties. As noted in the video, PLA makes for a very poor spring material, so you probably want to skip that one.

We covered compliant mechanisms in the past for purposes like blasters, including some that you can only see under a microscope.

Continue reading “The Challenges Of 3D Printing Reliable Springs” →

Trying Pair Programming With An LLM Chatbot

April 27, 2026 by Maya Posch 73 Comments

When it comes to software developers, there are a few distinct types. For example, the extroverted, chatty type, who is always going out there to share the latest and newest libraries and projects with everyone, and is very much into bouncing ideas off others, regardless of whether they know what you’re talking about. Then there is the introverted loner, who prefers to tackle programming challenges by bouncing things around inside their own minds and going on long walks to mull things over before committing to anything significant.

This leads to interesting scenarios when it comes to management-enforced ‘optimization’ strategies, like Pair Programming. This approach involves two developers sharing the same computer and keyboard, theoretically doubling the effective output by some kind of metric, but realistically often leading to at least one side feeling pretty miserable and disconnected unless you put two of the chatty types together.

As a certified introverted loner developer, the idea of using an LLM chatbot as a coding assistant naturally triggers unpleasant flashbacks to hours of forced awkward pair ‘programming’. However, maybe using an LLM chatbot could be more pleasant because you can skip the whole awkward socializing bit. In order to give it a shake, I put together a little experiment to see whether LLM-based coding assistants is something that I could come to appreciate, unlike pair programming.

Continue reading “Trying Pair Programming With An LLM Chatbot” →