Large Language Models (LLMs) can produce extremely human-like communication, but their inner workings are something of a mystery. Not a mystery in the sense that we don’t know how an LLM works, but a mystery in the sense that the exact process of turning a particular input into a particular output is something of a black box.

This “black box” trait is common to neural networks in general, and LLMs are very deep neural networks. It is not really possible to explain precisely why a specific input produces a particular output, and not something else.

Why? Because neural networks are neither databases, nor lookup tables. In a neural network, discrete activation of neurons cannot be meaningfully mapped to specific concepts or words. The connections are complex, numerous, and multidimensional to the point that trying to tease out their relationships in any straightforward way simply does not make sense.

Neural Networks are a Black Box

In a way, this shouldn’t be surprising. After all, the entire umbrella of “AI” is about using software to solve the sorts of problems humans are in general not good at figuring out how to write a program to solve. It’s maybe no wonder that the end product has some level of inscrutability.

This isn’t what most of us expect from software, but as humans we can relate to the black box aspect more than we might realize. Take, for example, the process of elegantly translating a phrase from one language to another.

I’d like to use as an example of this an idea from an article by Lance Fortnow in Quanta magazine about the ubiquity of computation in our world. Lance asks us to imagine a woman named Sophie who grew up speaking French and English and works as a translator. Sophie can easily take any English text and produce a sentence of equivalent meaning in French. Sophie’s brain follows some kind of process to perform this conversion, but Sophie likely doesn’t understand the entire process. She might not even think of it as a process at all. It’s something that just happens. Sophie, like most of us, is intimately familiar with black box functionality.

The difference is that while many of us (perhaps grudgingly) accept this aspect of our own existence, we are understandably dissatisfied with it as a feature of our software. New research has made progress towards changing this.

Identifying Conceptual Features in Language Models

We know perfectly well how LLMs work, but that doesn’t help us pick apart individual transactions. Opening the black box while it’s working yields only a mess of discrete neural activations that cannot be meaningfully mapped to particular concepts, words, or whatever else. Until now, that is.

Recent developments have made the black box much less opaque, thanks to tools that can map and visualize LLM internal states during computation. This creates a conceptual snapshot of what the LLM is — for lack of a better term — thinking in the process of putting together its response to a prompt.

Anthropic have recently shared details on their success in mapping the mind of their Claude 3.0 Sonnet model by finding a way to match patterns of neuron activations to concrete, human-understandable concepts called features.

A feature can be just about anything; a person, a place, an object, or more abstract things like the idea of upper case, or function calls. The existence of a feature being activated does not mean it factors directly into the output, but it does mean it played some role in the road the output took.

With a way to map groups of activations to features — a significant engineering challenge — one can meaningfully interpret the contents of the black box. It is also possible to measure a sort of relational “distance” between features, and therefore get an even better idea of what a given state of neural activation represents in conceptual terms.

Making Sense of it all

One way this can be used is to produce a heat map that highlights how heavily different features were involved in Claude’s responses. Artificially manipulating the weighting of different concepts changes Claude’s responses in predictable ways (video), demonstrating that the features are indeed reasonably accurate representations of the LLM’s internal state. More details on this process are available in the paper Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet.

Mapping the mind of a state-of-the-art LLM like Claude may be a nontrivial undertaking, but that doesn’t mean the process is entirely the domain of tech companies with loads of resources. Inspectus by [labml.ai] is a visualization tool that works similarly to provide insight into the behavior of LLMs during processing. There is a tutorial on using it with a GPT-2 model, but don’t let that turn you off. GPT-2 may be older, but it is still relevant.

Research like this offers new ways to understand (and potentially manipulate, or fine-tune) these powerful tools., making LLMs more transparent and more useful, especially in applications where lack of operational clarity is hard to accept.

[Sophia Dai] brings us a project you will definitely like if you’re tired of traditional peripherals like a typical keyboard and mouse combo. This is ErgO, a smart ring you can build out of a few commonly available breakouts, and it keeps a large number of features within a finger’s reach. The project has got an IMU, a Pimoroni trackball, and a good few buttons to perform actions or switch modes, and it’s powered by a tiny Bluetooth-enabled devboard so it can seamlessly perform HID device duty.

While the hardware itself appears to be in a relatively early state, there’s no shortage of features, and the whole experience looks quite polished. Want to lay back in your chair yet keep scrolling the web, clicking through links as you go? This ring lets you do that, no need to hold your mouse anymore, and you can even use it while exercising. Want to do some quick text editing on the fly? That’s also available; the ErgO is designed to be used for day to day tasks, and the UX is thought out well. Want to use it with more than just your computer? There is a device switching feature. The build instructions are quite respectable, too – you can absolutely build one like this yourself, just get a few breakouts, a small battery, some 3D printed parts, and find an evening to solder them all together. All code is on GitHub, just like you would expect from a hack well done.

Looking for a different sort of ring? We’ve recently featured a hackable cheap smart ring usable for fitness tracking – this one is a product that’s still being reverse-engineered, but it’s alright if you’re okay with only having an accelerometer and a few optical sensors.

Helium-neon lasers may be little more than glorified neon signs, but there’s just something about that glowing glass tube that makes the whole process of stimulated emission easier to understand. But to make things even clearer, you might want to take a step inside the laser with something like [Les Wright]’s open-cavity He-Ne laser.

In most gas lasers, the stimulated emission action takes place within a closed optical cavity, typically formed by a glass tube whose ends are sealed with mirrors, one of which is partially silvered. The gas in the tube is stimulated, by an electrical discharge in the case of a helium-neon laser, and the stimulated photons bounce back and forth between the mirrors until some finally blast out through the partial mirror to form a coherent, monochromatic laser beam. By contrast, an open-cavity laser has a gas-discharge tube sealed with the fully silvered mirror on one end and a Brewster window on the other, which is a very flat piece of glass set at a steep angle to the long axis of the tube and transparent to p-polarized light. A second mirror is positioned opposite the Brewster window and aligned to create a resonant optical cavity external to the tube.

To switch mirrors easily, [Les] crafted a rotating turret mount for six different mirrors. The turret fits in a standard optical bench mirror mount, which lets him precisely align the mirror in two dimensions. He also built a quick alignment jig, as well as a safety enclosure to protect the delicate laser tube. The tube is connected to a high-voltage supply and after a little tweaking the open cavity starts to lase. [Les] could extend the cavity to almost half a meter, although even a waft of smoke was enough obstruction to kill the lasing at that length.

If this open-cavity laser arrangement seems familiar, it might be because [Les] previously looked at an old-school particle counter with such a laser at its heart.

At Hackaday, we comb the world of tech in search of good things to bring you. Today’s search brought up something very familiar, [Jazzy Jane] has an Advance E1 tube signal generator, the same model as the unit on the shelf above where this is being written. It’s new to her, so she’s giving it a teardown and fixing any safety issues before powering it on.

For a 70+ year old unit, the quality of these instruments was such that they remain useful and reliable to this day. Unsurprisingly a few things need looking at, such as an aged mains lead and a pair of filter caps in the power supply which haven’t aged well. These parts failed on the E1 here too, and while she’s taking the time to order appropriate replacements we have to admit to being cheapskates and robbing parts with an appropriate working voltage for ours from a nearby PC power supply.

Where this one becomes rather interesting is in an extra switch and socket. It’s a wafer switch with a load of capacitors, and the best guess is it provides some adjustability for the inbuilt audio oscillator which had a fixed frequency on stock models. This is part one of a series though, so we’re looking forward to finding out its purpose in the next installment. Take a look at the video below the break, and if that’s not enough, we seem to have had more than one piece of vintage British test equipment here of late.

USB drives are incredibly useful, both storing files for transport between different computers and for creating bootable drives that let us use or install other operating systems on our computers. While online file storage systems like Dropbox and Google Drive have taken over a large percentage of the former task from USB drives, they have not been able to act as bootable media, ensuring that each of us have a few jump drives lying around. That might not be the case anymore, though, as this guide is the first we know of to be able to use Google Drive to boot to a Linux system.

Unlike the tried-and-true jump drive methods, however, this process is not straightforward at all. It relies on two keys, the first of which is FUSE which allows a filesystem to be created in userspace. The second is exploiting a step in boot process of Linux systems where the kernel unpacks a temporary filesystem, called initramfs, in order to load the real filesystem. Normally a user doesn’t interact much with this step, but that doesn’t mean it’s impossible. A tool called dracut allows using an existing Linux installation to build a custom initramfs and in this case, the custom initramfs is built to include the proper support for both networking and FUSE.

The proof of concept in this demonstration originally ran in a container, using an existing project called google-drive-ocamlfuse to interact with Google Drive itself. From there, after sorting out some issues with root access, networking, malfunctioning symlinks, and various timeouts on the (perhaps predictably) slow system, the whole contraption was moved over to a laptop so it could be tested on real hardware. Everything runs, and although the original creator of this behemoth admits it is a bit “silly” they note that there may be some real-world use cases for something like this. We still won’t expect everyone to throw out their jump drives anytime soon, though. If you’re not feeling like your Linux skills are up to the challenge of something like this, we’d recommend you start with our own [Al Williams]’s Linux Fu series.

Most 3D printer filament soaks up water from the air, and when it does, the water passing through the extruder nozzle can expand, bubble, and pop, causing all kinds of mayhem and unwanted effects in the print. This is why reels come vacuum sealed. Some people 3D print so much that they consume a full roll before it can soak up water and start to display these effects. Others live in dry climates and don’t have to worry about humidity. But the rest of us require a solution. To date, that solution has been filament dryers, which are heated elements in a small reel-sized box, or for the adventurous an oven put at a very specific temperature until the reel melts and coats the inside of the oven. The downside to this method is that it’s a broad stroke that takes many hours to accomplish, and it’s inefficient because one may not use the whole roll before it gets soaked again.

In much the same way that instant water heaters exist to eliminate the need for a water heater, [3DPI67] has a solution to this problem, and it involves passing the filament through a small chamber with a heating element and fan circulating air. The length of the chamber is important, as is the printing speed, since the filament needs to have enough time in the improvised sauna to sweat out all its water weight. The temperature of the chamber can’t get above the glass transition temperature of the filament, either, which is another limiting factor for the dryer. [3DPI67] wrote up a small article on his improvised instant filament heater in addition to the video.

So far, only TPU has been tested with this method, but it looks promising. Some have suggested a larger chamber with loops of filament so that more can be exposed for longer. There’s lots of room for innovation, and it seems some math might be in order to determine the limits and optimizations of this method, but we’re excited to see the results.

It’s official: [Engineer Bo] wins the internet with a video titled “Finding NEMA 17,” wherein he builds a dynamometer to find the best stepper motor in the popular NEMA 17 frame size.

Like a lot of subjective questions, the only correct answer to which stepper is best is, “It depends,” and [Bo] certainly has that in mind while gathering the data needed to construct torque-speed curves for five samples of NEMA 17 motors using his homebrew dyno. The dyno itself is pretty cool, with a bicycle disc brake to provide drag, a load cell to measure braking force, and an optical encoder to measure the rotation of the motor under test. The selected motors represent a cross-section of what’s commonly available today, some of which appear in big-name 3D printers and other common applications.

[Bo] tested each motor with two different drivers: the TMC2209 silent driver to start with, and because he released the Magic Smoke from those, the higher current TB6600 module. The difference between the two drivers was striking, with lower torque and top speeds for the same settings on each motor using the TB6600, as well as more variability in the data. Motors did better across the board with the TBC6600 at 24 volts, showing improved torque at higher speeds, and slightly higher top speeds. He also tested the effect of microstepping on torque using the TBC6600 and found that using full steps resulted in higher torque across a greater speed range.

At the end of the day, it seems as if these tests say more about the driver than they do about any of the motors tested. Perhaps the lesson here is to match the motor to the driver in light of what the application will be. Regardless, it’s a nice piece of work, and we really appreciate the dyno design to boot — reminds us of a scaled-down version of the one [Jeremey Fielding] demonstrated a few years back.

[MakerM0]’s LangCard is an entry into our 2024 Business Card Challenge that just so happens to fit the Keebin’ bill as well.

You might label this a pocket cyberdeck, and that’s just fine with me. The idea here is to have a full-keyboard development board for learning programming languages like CircuitPython, MicroPython, C++, and so on, wherever [MakerM0] happens to be at a given moment.

Open up the LangCard and you’ll find an RP2040 and a slim LiPo battery. I’m not sure what display that is, but there are probably a few that would work just fine were you to make one of these fun learning devices for yourself.

Calling All Tiny Keyboard Makers!

It seems that [sporewoh], who has been featured here before for building magnificent tiny keyboards, is holding a tiny keyboard design contest, which is being sponsored by PCBWay.

All the rules and such are available over on GitHub. Basically, you need to create a new design, publish the open-source design somewhere along with the source files, and, ideally, build a functional prototype. Entries are made official by sharing in the appropriate channel of [sporewoh]’s Discord.

Entries will be scored on novelty and innovation, size/portability, viability, reproducibility, and presentation. Submissions aren’t due until September 10th, so you have a bit of time to really think about what you’re going to do. The prizes include PCBWay credits as well as kits designed by [sporewoh]. How small can you go and still be able to type at least 20 WPM? That’s a requirement, by the way.

The Centerfold: It’s a Tasty Snacks Board

That’s right; we’ve got a gallery this time. I simply couldn’t decide which picture best conveyed the deliciousness of this thing. I mean, the first shot is a good thought, but you really don’t get right away that it’s a Ring Pop knob, and that’s vitally important information.

Anyway, this is a Le Chiffre that has been quite smoothly 3D-printed in marble filament. [CattiDaddi] says they got that by using a matte print bed. I wish I knew what keycaps those are, because that is a sneaky typeface they have going.

Do you rock a sweet set of peripherals on a screamin’ desk pad? Send me a picture along with your handle and all the gory details, and you could be featured here!

Historical Clackers: the Forgotten History of Chinese Keyboards

Here is quite an interesting bit of history as it relates to China’s ability to survive the ravages of time, which pivots on their use of character-based script.

The story begins with a talk that took place decades ago, and follows the path of one audience member who came to change the course of Chinese keyboard history — a Taiwanese cadet named Chan-hui Yeh.

After graduating with a B.S. in electrical engineering, Yeh went on to earn an M.S. in nuclear engineering and a Ph.D. in electrical engineering. He then joined IBM, although it wasn’t to revolutionize Chinese text technologies; he was helping to develop computational simulations for large-scale manufacturing plants. But the talk stuck with him.

Yeh eventually quit his job and developed the IPX keyboard, which had 160 main keys with 15 characters each. A daughter keyboard was used to choose the character on a given key, and there were nearly 120 levels of “Shift” to change all the 160 keys’ character assignments.

See that picture with the spiral-bound book? The 160 keys are underneath the book, and the user presses the pages to access the pressure pads beneath. The booklets had up to eight pages, each with 2,400 characters. The total number of potential symbols was just under 20,000.

The IPX keyboard is just the first of three interesting inputs described within this history. You owe it to yourself to devote time to reading this one.

Thanks to [juju] for sending this in!

ICYMI: One-Handed Keyboard Does It Without Chording

Usually when we talk about operating an entire keyboard with one hand and leaving the other free for mousing or holding a beer, chording — pressing multiple keys at once like on a piano — is very much on the table. Keyboards like the Infogrip BAT come to mind.

But that isn’t always the case. Take for example the one-handed PCD Maltron, which I think must have inspired [Dylan Turner]’s one-handed keyboard.

[Dylan]’s design puts 75 keys in close reach of one hand’s worth of fingers, and doesn’t let the thumb off easy like on a standard keyboard. All the Function keys are  there, and the arrow keys are in a familiar layout. There is even an Insert/Delete cluster. Everything is up on GitHub if you want to make your own.

---

Got a hot tip that has like, anything to do with keyboards? Help me out by sending in a link or two. Don’t want all the Hackaday scribes to see it? Feel free to email me directly.

In the monthly Ladybird Browser update video which we’ve placed below, SerenityOS founder [Andreas Kling] announced an interesting development. The browser has been forked from the OS that has been its progenitor, and both projects will now proceed separately. This frees the browser from the SerenityOS insistence on avoiding external libraries, and allows it to take advantage of stable, fast, and mature open source alternatives. This is already paying dividends in compatibility and speed, and is likely to lead further towards a usable everyday browser as time goes by.

As the world of fully-featured web browser engines has contracted from a number of different projects to little more than Google’s Blink and Mozilla’s Gecko, Ladybird has found itself in an unexpected position. It is vital that the browser market retains some competition and does not become a Google monoculture, so while it might not seem so at first glance, the news of Ladybird going alone has the potential to be one of the most far-reaching open source stories of the year.

If you’d like to try Ladybird you’ll have to get your hands slightly dirty and build it yourself, but we’d expect ready-built versions to appear in due course. We took a look at an earlier version of Ladybird last year, as well as SerenityOS itself.

[ZXGuesser] has pulled off a true feat of Meccano engineering: building a Meccano Hellschreiber machine. The design is a close replica of the original Siemens Feld-Hell machine as documented here. What is Hellschreiber, you might ask? It’s a very neat method of sending written messages over the air by synchronizing a printing wheel on the receiving end with pulses generated on the transmitter. By quickly moving the print wheel up and down, arbitrary figures can be printed out. If you want to learn more about Hellschreiber, check out this excellent Hackaday post from almost a decade ago!

The Mastodon thread linked above goes into more detail about the difficulty in building this behemoth — and the slight regret of sticking with the authentic QWERTZ keyboard layout! In order to use the Hellschreiber mode, you have to keep up a steady rhythm of typing at about 2.5 characters per second, otherwise, the receiving end will see randomly spaced gaps between each letter. So while having to type at a steady speed [ZXGuesser] also had to work with a slightly different keyboard layout. Despite this difficulty, some very good quality output was generated!

Incredibly, the output looks just like the output from the original, century-old design. We think this is an absolutely incredible accomplishment, and we hope [ZXGuesser] doesn’t follow through on disassembling this amazing replica — or if they do, we hope it’s documented well enough for others to try their hand at it!

Thanks [BB] for the tip!

In the days before integrated circuits became ubiquitous, providing advanced functionality in a single package, designers became adept at extracting the maximum use from discrete components. They’d use clever circuits in which a transistor or other active part would fulfill multiple roles at once, and often such circuits would need more than a little know-how to get working. It’s not often in 2024 that we encounter this style of circuit, but here’s [Maurycy] with a cheap microwave radar module doing just that.

On the board is an RF portion with a single transistor, some striplines, and an SOIC chip. Oddly this last part turns out to be an infra-red proximity sensor chip, so what’s going on? Careful analysis of the RF circuit reveals something clever. As expected, it’s a 3.18 GHz oscillator, but how is it functioning as both transmitter and receiver? The answer comes in the form of a resistor and capacitor in the emitter circuit, which causes the transistor to also oscillate at about 20 kHz. The result is that at different times in the 20 kHz period, the transistor is either off, fully oscillating at 3.18GHz and transmitting, or briefly in the not-quite-oscillating state between the two during which it functions as a super-regenerative receiver. This is enough for one device to effectively transmit and receive at the same time with the minimum of parts, there’s no need for a mixer diode as you might expect if it were it a direct conversion receiver. Perhaps in RF terms, it’s not particularly pretty, but we have to admit to being impressed by its simplicity. He goes on to perform a few experiments with the board as a transmitter or as a more conventional radar.

This isn’t the first such radar module we’ve looked at, here’s one designed from scratch. And we love regens, since they are so simple to build.

One of the fun parts of the ESP32-S3 microcontroller is that it got upgraded to the newer Cadence Xtensa LX7 processor core, which turns out to have a range of SIMD instructions that can help to significantly speed up a range of tasks. [Shranav Palakurthi] recently used this to speed up the processing of video frames to detect corners using the FAST method. By moving some operations that benefit from SIMD over to an optimized version written in LX7 ASM, the algorithm’s throughput was increased by 220%, from 5.1 MP/s to 11.2 MP/s, albeit with some caveats.

The problem with the SIMD instructions in the LX7 other than them being very poorly documented – unless you sign an NDA with Cadence –  is that it misses many instructions that would be really useful. For [Shranav] the lack of support for direct misaligned reads and comparing of unsigned 8-bit numbers were hurdles, but could be worked around, with the results available on GitHub.

Much of the groundwork for this SIMD implementation was laid by [Larry Bank], who reverse-engineered the SIMD instructions from available documentation and code samples, finding that the ESP32-S3 misses quite a few common SIMD instructions, including various shifts and unaligned reads and writes. Still, it’s good enough for quite a few tasks, as long as you can make it work with the available instructions.

[mircemk] shows how to create a simple non-contact proximity sensor using little more than an Arduino Nano board, and a convenient software library intended to measure the value of capacitors.

The basic idea is that it’s possible to measure a capacitor’s capacitance using two microcontroller pins and the right software, so by using a few materials to create an open-style capacitor, one can monitor it for changes and detect when anything approaches enough to alter its values past a given threshold, creating a proximity sensor.

The sensor shown here is essentially two plates mounted side-by-side, attached to an Arduino Nano using the Capacitor library which uses just two pins, one digital and one analog.

As configured, [mircemk]’s sensor measures roughly thirty picofarads, and that value decreases when approached by something with a dielectric constant that is different enough from the air surrounding the sensor. The sensor ignores wood and plastic, but an approaching hand is easily detected. The sensor also detects liquid water with similar ease, either in the form of pooled liquid, or filled bottles.

We’ve also seen a spring elegantly used as a hidden touch sensor that works through an enclosure’s wall by using similar principles, so the next time you need a proximity or touch-sensitive sensor in a project, reaching for the junk box might get you where you need to go. Watch [mircemk]’s sensor in action in the video, just below the page break.

There’s no shortage of USB-C chargers in all sorts of configurations, but sometimes, you simply need a few more charging ports on the go, and you got a single one. Well then, check out [bluepylons]’s USB-C splitter, which takes a single USB-C 5V/3A port and splits it into three 5V/1A plugs, wonderful for charging a good few devices on the go!

This adapter does things right – it actually checks that 3A is provided, with just a comparator, and uses that to switch power to the three outputs, correctly signalling to the consumer devices that they may consume about 1A from the plugs. This hack’s documentation is super considerate – you get detailed instructions on how to reproduce it, every nuance you might want to keep in mind, and even different case options depending on whether you want to pot the case or instead use a thermal pad for a specific component which might have to dissipate some heat during operation!

This hack has been documented with notable care for whoever might want to walk the journey of building one for themselves, so if you ever need a splitter, this one is a wonderful weekend project you are sure to complete. Wonder what kind of project would be a polar opposite, but in all the best ways? Why, this 2kW USB-PD PSU, most certainly.

Conservation of energy isn’t just a good idea: It is the law. In particular, it is the first law of thermodynamics. But, apparently, a lot of people don’t really get that because history is replete with inventions that purport to run forever or produce more energy than they consume. Sometimes these are hoaxes, and sometimes they are frauds. We expect sometimes they are also simple misunderstandings.

We thought about this when we ran across the viral photo of an EV with a generator connected to the back wheel. Of course, EVs and hybrids do try to reclaim power through regenerative braking, but that’s recovering a fraction of the energy already spent. You can never pull more power out than you put in, and, in fact, you’ll pull out substantially less.

Not a New Problem

If you think this is a scourge of social media and modern vehicles, you’d be wrong. Leonardo da Vinci, back in 1494, said:

Oh ye seekers after perpetual motion, how many vain chimeras have you pursued? Go and take your place with the alchemists.

There was a rumor in the 8th century that someone built a “magic wheel,” but this appears to be little more than a myth. An Indian mathematician also claimed to have a wheel that would run forever, but there’s little proof of that, either. It was probably an overbalanced wheel where the wheel spins due to weight and gravity with enough force to keep the wheel spinning.

An architect named Villard de Honnecourt drew an impractical perpetual motion machine in the 13th century that was also an overbalanced wheel. His device, and other similar ones, would require a complete lack of friction to work. Even Leonardo da Vinci, who did not think such a device was possible, did some sketches of overbalanced wheels, hoping to find a solution.

Types of Machines

There isn’t just a single kind of perpetual motion machine. A type I machine claims to produce work without any input energy. For example, a wheel that spins for no reason would be a type I machine.

Type II machines violate the second law of thermodynamics. For example, the “zeromoter” — developed in the 1800s by John Gamgee, used ammonia and a piston to move by boiling and cooling ammonia. While the machine was, of course, debunked, Gamgee has the honor of being the inventor of the world’s first mechanically frozen ice rink in 1844.

Type III machines claim to use some means to reduce friction to zero to allow a machine to work that would otherwise run down. For example, you can make a flywheel with very low friction bearings, and with no load, it may spin for years. However, it will still spin down.

Often, machines that claim to be perpetual either don’t really last forever — like the flywheel — or they actually draw power from an unintended source. For example, in 1760, James Cox and John Joseph Merlin developed Cox’s timepiece and claimed it ran perpetually. However, it actually drew power from changes in barometric pressure.

Frauds

These inventions were often mere frauds. E.P. Willis in 1870 made money from his machine but it actually had a hidden source of power. So did John Ernst Worrell Keely’s induction resonance motion motor that actually used hidden air pressure tubes to power itself. Harry Perrigo, an MIT graduate, also demonstrated a perpetual motion machine to the US Congress in 1917. That device had a secret battery.

However, some inventors probably weren’t frauds. Nikola Tesla was certainly a smart guy. He claimed to have found a principle that would allow for the construction of a Type II perpetual motion machine. However, he never built it.

There have been hosts of others, and it isn’t always clear who really thought they had a good idea and how many were just out to make a buck. But some people have created machines as a joke. Dave Jones, in 1981, created a bicycle wheel in a clear container that never stopped spinning. But he always said it was a fake and that he had built it as a joke. Adam Savage looks at that machine in the video below. He wrote his secret in a sealed envelope before he died, and supposedly, only two people know how it works.

Methods

Most perpetual machines try to use force from magnets. Gravity is also a popular agent of action. Other machines depend on buoyancy (like the one in the video below) or gas expansion and condensation.

The US Patent and Trademark Office manual of patent examining practice says:

With the exception of cases involving perpetual motion, a model is not ordinarily required by the Office to demonstrate the operability of a device. If operability of a device is questioned, the applicant must establish it to the satisfaction of the examiner, but he or she may choose his or her own way of so doing.

The UK Patent Office also forbids perpetual motion machine patents. The European Patent Classification system has classes for “alleged perpetua mobilia”

Of course, having a patent doesn’t mean something works; it just means the patent office thought it was original and can’t figure out why it wouldn’t work. Consider Tom Bearden’s motionless electromagnetic generator, which claims to generate power without any external input. Despite widespread denouncement of the supposed operating principle — Bearden claimed the device extracted vacuum energy — the patent office issued a patent in 2002.

The Most Insidious

The best machines are ones that use energy from some source that isn’t apparent. For example, a Crookes radiometer looks like a lightbulb with a little propeller inside. Light makes it move. It is also a common method to use magnetic fields to move something without obviously spinning it. For example, the egg of Columbus (see the video below) is a magnet, and a moving magnetic field makes the egg spin. This isn’t dissimilar from a sealed pump where a magnet turns on the dry side and moves the impeller, which is totally immersed in liquid.

Some low-friction systems, like the flywheel, can seem to be perpetual motion machines if you aren’t patient enough. But eventually, they all wear down.

Crazy or Conspiracy?

Venues like YouTube are full of people claiming to have free energy devices that also claim to be suppressed by “the establishment”. While we hate to be on the wrong side of history if someone does pull it off, we are going to go out on a limb and say that there can’t be a true perpetual motion machine. Unless you cheat, of course.

This is the place we usually tell you to get hacking and come up with something cool. But, sadly, for this time we’ll entreat you to spend your time on something more productive, like a useless box or put Linux on your Commodore 64.

 

There’s a mystique in old keyboard circles around the IBM Model M, the granddaddy of PC keyboards with those famous buckling spring key switches. The original Model M was a substantial affair with a sheet metal backplane that would probably serve well as a weapon in a zombie apocalypse and still allow writing a Hackaday piece afterward, but later on in the life of these ‘boards there was also a lighter version. The M2 as these models are dubbed has a few known problems, and [Anders Nielsen] scored one online that turned out to have dodgy capacitors. His video, below the break, takes us through the disassembly of his M2 and provides a relaxing tour of these not-quite-so-famous peripherals.

As you’d expect, three-decade-old plastic isn’t always in the best shape, so disassembly and unlatching all those little tabs has to be performed with care. The keys come off and the springs are on show, but we get a nasty shock when they all fall out of place as the top is removed. It appears the rookie mistake is to not turn the ‘board upside down before parting it. Replacing the caps is an easy process after all that, and we get a little dive into the 6805 processors used in model Ms.

If you have a model M of any description then you’re probably at home with the clack-clack-clack sound they make, but have you ever looked at its ancestor, the model F?

In the communications surrounding LLMs and popular interfaces like ChatGPT the term ‘hallucination’ is often used to reference false statements made in the output of these models. This infers that there is some coherency and an attempt by the LLM to be both cognizant of the truth, while also suffering moments of (mild) insanity. The LLM thus effectively is treated like a young child or a person suffering from disorders like Alzheimer’s, giving it agency in the process. That this is utter nonsense and patently incorrect is the subject of a treatise by [Michael Townsen Hicks] and colleagues, as published in Ethics and Information Technology.

Much of the distinction lies in the difference between a lie and bullshit, as so eloquently described in [Harry G. Frankfurt]’s 1986 essay and 2005 book On Bullshit. Whereas a lie is intended to deceive and cover up the truth, bullshitting is done with no regard for, or connection with, the truth. The bullshitting is only intended to serve the immediate situation, reminiscent of the worst of sound bite culture.

When we consider the way that LLMs work, with the input query used to provide a probability fit across the weighted nodes that make up its vector space, we can see that the generated output is effectively that of an oversized word prediction algorithm. This precludes any possibility of intelligence and thus cognitive awareness of ‘truth’. Meaning that even if there is no intent behind the LLM, it’s still bullshitting, even if it’s the soft (unintentional) kind. When taking into account the agency and intentions of those who created the LLM, trained it, and created the interface (like ChatGPT), however, we enter into hard, intentional bullshit territory.

It is incidentally this same bullshitting that has led to LLMs being partially phased out already, with Retrieval Augmented Generation (RAG) turning a word prediction algorithm into more of a fancy search machine. Even venture capitalists can only take so much bullshit, after all.

A couple of weeks back we featured a story (third item) about a chunk of space jetsam that tried to peacefully return to Earth, only to find a Florida family’s roof rudely in the way. The 700-gram cylinder of Inconel was all that was left of a 2,360-kg battery pack that was tossed overboard from the ISS back in 2021, the rest presumably turning into air pollution just as NASA had planned. But the surviving bit was a “Golden BB” that managed to slam through the roof and do a fair amount of damage. At the time it happened, the Otero family was just looking for NASA to cover the cost of repairs, but now they’re looking for a little more consideration. A lawsuit filed by their attorney seeks $80,000 to cover the cost of repairs as well as compensation for the “stress and impact” of the event. This also seems to be about setting a precedent, since the Space Liability Convention, an agreement to which the USA is party, would require the space agency to cover damages if the debris had done damage in another country. The Oteros think the SLC should apply to US properties as well, and while we can see their point, we’d advise them not to hold their breath. We suppose something like this had to happen eventually, and somehow we’re not surprised to see “Florida Man” in the headlines.

There was a little hubbub this week around the release of a study regarding the safety of autonomous vehicles relative to their meat-piloted counterparts. The headlines for the articles covering this varied widely and hilariously, ranging from autonomous vehicles only being able to drive in straight lines to AVs being safer than human-driven cars, full-stop. As always, one has to read past the headlines to get an idea of what’s really going on, or perhaps even brave reading the primary literature. From our reading of the abstract, it seems like the story is more nuanced. According to an analysis of crashes involving 35,000 human-driven vehicles and 2,100 vehicles with some level of automation, AVs with SAE Level 4 automation suffered fewer accidents across the board than those without any automation. Importantly, the accidents that Level 4 vehicles do suffer are more likely to occur when the vehicle is turning just before the accident, or during low-visibility conditions such as dawn or dusk. The study also compares Level 4 automation to Level 2, which has driver assistance features like lane-keeping and adaptive cruise control, and found that Level 2 actually beats Level 4 in clear driving conditions, but loses in rainy conditions and pretty much every other driving situation.

There’s a strange story coming out of New York regarding a Federal Communications Commission (FCC) enforcement action that seems a little shady. It regards a General Mobile Radio Service (GMRS) repeater system used by the New York State GMRS Alliance. GMRS is sort of a “ham radio lite” system — there’s no testing required for a license, you just pay a fee — that uses the UHF band. Repeaters are allowed, but only under specific rules, and that appears to be where things have gone wrong for the club. The repeater system they used was a linked system, which connected geographically remote repeaters stretching from the far western part of the state near Buffalo all the way to Utica. It’s the linking that seems to have raised the FCC’s hackles, and understandably so because it seems to run counter to the GMRS rules in section 95. But it’s the method of notification that seems hinky here, as the repeater custodian was contacted by email. That’s not typical behavior for the FCC, who generally send enforcement notices by certified snail mail, or just dispense with the paper altogether and knock on your door. People seem to think this is all fake news, and it may well be, but then again, the email could just have been an informal heads-up preceding a formal notice. Either way, it’s bad news for the GMRS fans in upstate New York who used this system to keep in touch along Interstate 90, a long and lonely stretch of road that we know all too well.

Third time’s a charm? We’ll see when sunspot region AR3723 (née AR3697 née AR3664) makes a historic third pass around the Sun and potentially puts Earth in its crosshairs yet again. The region kicked up quite a ruckus on its first pass across the solar disk back in May with a series of X-class flares that produced stunning aurorae across almost all of North America. Pass number two saw the renamed region pass more or less quietly by, although it did launch an M-class flare on June 23 that caused radio blackouts in most of the North Atlantic basin. When AR3723 does peek out from behind the eastern limb of the Sun it’ll be a much-diminished version of its former Carrington-level glory, and will likely be given multiple designations thanks to fragmentation while it was hanging out on the backside. But it could still pack a punch, and even if this particular region doesn’t have much juice left, it sure seems like the Sun has plenty of surprises in store for the balance of Solar Cycle 25.

Somebody made a version of Conway’s Game of Life using nothing but checkboxes, which is very cool and you should check it out.

And finally, we’ve been doing an unexpected amount of automotive DIY repairs these days, meaning we spend a lot of time trolling around for parts. Here’s something we didn’t expect to see offered by a national retailer, but that we’d love to find a use for. If it ever comes back in stock we just might pick one up.

You might think Google Glass was an innovative idea, but [Allison Marsh] points out that artist [Lisa Krohn] imagined the Cyberdesk in 1993. Despite having desk in the name, the imagined prototype was really a wearable computer. Of course, in 1993, the technology wasn’t there to actually build it, but it does look like [Krohn] predicted headgear that would augment your experience.

Unlike Google Glass, the Cyberdesk was worn like a necklace. There are five disk-like parts that form a four-key keyboard and something akin to a trackpad. There were two models built, but since they were nonfunctional, they could have any imagined feature you might like. For example, the system was supposed to draw power from the sun and your body, something practical devices today don’t really do, either.

She also imagined a wrist-mounted computer with satellite navigation, a phone, and more. Then again, so did [Chester Gould] when he created Dick Tracy. The post also talks about a more modern reimagining of the Cyberdesk last year.

While this wasn’t a practical device, it is a great example of how people imagine the future. Sometimes, they miss the mark, but even then, speculative art and fiction can serve as goals for scientists and engineers who build the actual devices of the future.

We usually think about machines augmenting our intelligence and senses, but maybe we should consider more physical augmentation. We do appreciate seeing designs that are both artistic and functional.