Large language models (LLMs) output language, but they are commonly tasked with making a decision or classification of some kind instead of writing an essay or chat reply. An LLM will be provided with input, and asked to classify that content in some way: with a rating, yes/no answer, a best-fit categorization, and so forth. A recent new type of model by the name of Jev was released only weeks ago and it is extremely fast, ultra-cheap, and laser-focused on that decision-making role. It can’t write even a single sentence, but it can classify and categorize very, very quickly.

Jev works like this: it still accepts text input, but it outputs only floating-point numbers. Those numbers are the “answers” to user-specified yes/no type questions, lists of choices, and scoring-type requests. [Simon Willison] provides a concise summary of what Jev does, and what makes this new category of model so interesting.

To say that the idea has caught on would be a wild understatement. Folks are making their own decision-type models and experiments in a flurry. Kev and Nimble are two examples (Nimble was added as a supported model in Ollama just recently, and is small enough to run locally with relative ease.)

If this type of local AI model was the missing link you needed to get an idea working, don’t keep it to yourself! Tell us all about it on the tips line.

There are some projects that seem at first sight to be easy, but anyone who tries them finds a whole heap of unexpected problems and turns to the off-the-shelf device. PCB antennas for example, or data links using LEDs, whether IR or visible. The latter doesn’t faze [Luca Soltoggio] though, because he has two ESP32s talking to each other using visible light. Best of all, both use a single LED as both transmitter and receiver.

The software is called SecurePair, and is an Arduino library for exchanging keys and communicating with encryption. The LEDs are the cool hardware hack but it’s designed to work with ESPNow or LoRa too, indeed a typical use case would see light for pairing and wireless for the exchange of encrypted data.. In case you were wondering, it relies on the property of an LED that it’s also a photodiode of sorts. Best of all, while the examples have two ESP32s, it’s not limited to that number and many more can join the conversation if needed.

Check out the video below to see it in action — if you’re curious about LEDs as sensors, we’ve been there too.

With the rise of synthetic polymers in everywhere from clothing, packaging and beyond, we have also seen a corresponding rise in fragments of these polymers in the environment. These micro- and nanoplastics (MNPs) come in a wide variety of sizes, but have in common that they do not break down very easily, leaving filtration as an important way to keep them out of our potable water sources. A recent paper by [Ethan A. Crawford] et al. in  Separation and Purification Technology details a way to use a low-cost, 3D-printed filtration mesh to filter out up to 90% of MNPs using their prototype.

This prototype uses a multi-layer filtration system printed in PLA, creating a flow-through system in which polyethylene glycol (PEG) acts as a sacrificial additive to the PLA. Whereas creating the fine pores required for filtration of MNPs is impossible using FDM printing, the spheres of PEG that form inside the printed filter can be subsequently etched away using nothing but hot (80°C) water, leaving behind a porous surface capable of trapping fine particles.

These pores are characterized in the paper, with the PLA-PEG10 sample showing the best porous structure while maintaining structural integrity of the PLA material. With 20 layers of these filters a filtration efficiency of 90% was achieved, though as the authors note aspects like improving the system and potential reusability still have to be investigated.

We have previously looked at MNPs, including how little we know about how many of them really are inside our bodies right now, and how lab gloves may be contaminating test results.

Building anything that flies from scratch is an ambitious undertaking, even if it’s ‘just’ radio-controlled. 3D printing the aircraft isn’t that odd these days, but putting a hot jet engine into a plastic airframe is another ambitous reach. Getting said 3D printed airframe up to the transonic speed of Mach 0.8? Ambition, thy name is Kingchaser. Or at least, that’s the name [The Mach Initiative] give to their very ambitious aircraft, the video about which is embedded below.

The airframe is largely 3D printed from PETG– that’s all the orange bits– but there are carbon fiber rods and an aluminum frame to help take up the strain. The black section around the motor is printed from PPS-CF– that’s Polyphenylene sulfide with carbon fibre–wrapped in carbon fiber to take the heat. This all builds off a smaller PLA version that’s already flown called Kingfisher, which claimed the title of the first 3D printed jet. That flight is on their YouTube channel, if you’re interested.

Now with a bigger aircraft and a much bigger engine– 300 N or 68 lbf of thrust– they’re going for the speed record. If they get even close to the design goal of mach 0.8, they’re absolutely going to leave the world’s fastest drone– a 626 km/h quadcopter we’ve written about— in the dust.

The switch to PETG from PLA, for the record, was to deal with the expected aerodynamic heating at that speed, about 990 km/h or 615 mph. An interesting detail many don’t bother with when it comes to 3D printed airplanes is that the skin has all been polished smooth, since skin drag is dominant at that speed regime. The video is just chock full of those little details that it takes to defeat drag and get to the record speed, and there’s more to come from [The Mach Initiative].

Speaking of remote controlled speed records, we covered another batch of brits take home the land speed record a few years back. May [The Mach Initiative] see such success!

Copper sulfate crystals are probably among one of the prettiest crystals you can grow at home, with even just a simple setup with some copper scrap and vinegar already capable of producing lots of them. Yet what if you want to grow really big ones? In that case the [Crystalverse] has got your back, with a recent video that expands on an older blog post.

In lieu of the copper-and-vinegar approach you can also obtain copper sulfate directly, since it’s a common fungicide, rootkiller as well as drying agent. This means that your local brick-and-mortar retailer or favorite online store probably has a few kg of the stuff available for sale.

As with most large crystal growing procedures the key is to have a saturated solution, which for copper sulfate just takes near-boiling water, to create a mesmerizingly blue liquid if there are no contaminants in it. By adding slightly more copper sulfate there are also crystallization sites on the bottom of the jar to draw these away from your large crystal.

From there it’s the same as with growing other large crystals – even those from sugar – with a seed crystal suspended into the solution, along with a silent prayer to the crystal gods that said seed crystal continues to grow without any defects. One gotcha with copper sulfate crystals is that the intense blue color is largely due to the presence of water molecules. This means that once it dehydrates, it turns effectively white.

Also of note that is the slower the crystal grows, the better the result is likely to be. During the months that it takes for these large crystals to grow, you need to carefully manage the copper sulfate solution, remove competing crystals on the bottom of the container and keep the temperature as constant as possible.

The discerning modern gamer’s quest is to find a quieter computer without sacrificing performance. That proves challenging when modern components are highly efficient at producing monumental quantities of thermal energy. So the typical solution ends up moving enough air through the computer to create a rather tumultuous racket. So why have fans at all? Well, as [Billet Labs] demonstrates, that’s not necessarily a good idea. 

The build begins with a block of aluminum which, with a bit of hand-metal work, may become the base of the PC “case.” A dry-fit of the aesthetically modified parts taken from a previous attempt at a fanless PC reveals the steampunk aesthetic that defines the build. On final assembly, a large amount of thermal paste under the flex ATX power supply and 6mm of thermal pads under the motherboard provide passive cooling to the block of aluminum they are attached to

The cooling loop begins with three radiators: a big one, a small one, and a slightly smaller one. But raw metal PC radiators are difficult and expensive to come by these days. Some paint stripping and polishing will fix that problem. A few hand-made mounting brackets complete the aesthetic. All three radiators are stacked above the motherboard, and on top of each other, positioned horizontally to encourage passive airflow through them.

At this point, a normal PC would be plumbed up using standard flexible or, if you’re particularly ambitious, hard piping. But this is no normal PC. Instead, the plumbing uses household copper piping, soldered together. Two gauges sit near the front to show the water pressure and temperature. The reservoir is a larger diameter pipe with one end soldered onto a smaller pipe leading to the pump. The other end is a screw cap, with the corners machined off to match the pipe, to allow for filling of the loop. Soldering this took quite some effort because of the thermal mass in such a large pipe.

Once testing commenced, two issues became quickly apparent. The pressure gauge creates quite the racket as water flows past, and the stacked radiators are inadequate at cooling the PC. Rather show-stopping issues when the goal is gaming in silence. Was all this work for naught? No, the design proved the block of aluminum quite adept at cooling the motherboard and GPU. Moreover, adding a simple 120mm fan between the two smaller radiators and removing the pressure gauge effectively quelled the overheating and ticking issue. The result is a quiet (and quite pretty) living room PC with performance to match.

Hollywood has a lot to answer for when it comes to “silencers” on firearms. On screen, a suppressed pistol makes a pinging thud, and the bad guy in the next room flops to the floor. Real life is rather less cinematic. A suppressor takes the edge off the report, but it’s by no means quiet or silent. Hence why the term “silencer” is more of a colloquialism, and why “suppressor” is more appropriate.

Every so often, though, someone gets remarkably close to the movie version. The best-known example is probably the De Lisle carbine. A British commando weapon from the Second World War, it was cobbled together from bits and pieces and ended up being a stellar performer in the world of clandestine operations.

Softly

William Godfray De Lisle was an engineer who worked at the Air Ministry during the period in which the United Kingdom found itself embroiled in the tumult that was World War II. He happened to be a capable weaponsmith, and during the war, he put together a rather unique carbine in .22 calibre as a personal project. Based on a short-magazine Lee-Enfield Mk III, it had the unique feature of having a sizable suppressor integrated into the very design of the weapon itself. The homebrew project was field tested on rabbits and small game at De Lisle’s property, with the spoils of the hunt proving most welcome at a time when heavy rationing was in place.

Soon enough, De Lisle was invited to demonstrate the weapon for examination by Major Sir Malcolm Campbell of Combined Operations HQ, and De Lisle’s design would prove remarkable with its performance in testing. The 1940s were rather a different time to today, without so many concerns around health, safety, or flinging bullets in the vicinity of the public. Thus, a simple trial was designed. The rifle was fired from the top of the Adelphi Building directly into the Thames River, and it was noted that passers by did not appear to hear it firing or respond to the shots at all. It was a ringing endorsement that the De Lisle design did what it was supposed to do.

The gun’s success was heavily down to the integrated suppressor. It was inspired by the basic Maxim design, made popular by Hiram Percy Maxim, son of the creator of the famed Maxim machine gun that was so popular in the previous World War. At two inches in diameter, it used two expansion chambers, with a spiral system of baffles inside to absorb and diffuse the energy of the shot, reducing the sound output at the source. By virtue of its large size, it was capable of accepting a high volume of gases when fired, aiding its performance. The suppressor was highly effective at both reducing noise and cutting muzzle flash, both of prime interest to commandos involved in covert operations.

As it stood, however, De Lisle’s carbine wasn’t quite ideal for wartime use in its prototype form. The .22 cartridge was plenty for hunting small game, but rather titchy for the job of taking down German sentries. It was thus requested that a 9 mm version be produced, against De Lisle’s advice.

The following prototype was unsuccessful both ballistically and in terms of the noise it made—in part due to the supersonic nature of the 9 mm ammunition. De Lisle instead pushed ahead with a redesign chambered for .45 ACP, which had the benefit of increased stopping power and subsonic muzzle velocity for a quieter report when fired.

It proved ideal for the job, and soon enough, the new ultra-quiet carbine was in the hands of commandos conducting missions in occupied France. While a handful of handmade pre-production units were already taking part in the war, official range test were further proving the value of the new gun. Where a contemporary Sten gun was hitting 125 dB, and a silenced version sounding off at 89.5 dB, the De Lisle was besting that with a 85.5 dB report even despite firing a much larger round. It was considered effective out to a range of approximately 200 meters.

The De Lisle carbine was soon enough given a manufacturing contract for 500 units, though it would never be produced in such numbers. The final count was just 130 examples, with the majority being delivered directly to Combined Operations HQ. Beyond the suppressor, the gun was still effectively assembled from bits and pieces even in its production form. The heart of the weapon was the short-magazine Lee-Enfield Mk III, donating its bolt-action mechanics, but it had been hacked to accept a modified Thompson submachine gun barrel suitable for the .45 ACP cartridge. The six-groove barrel was also ported to help control the venting of gases into the suppressor assembly, which stretched from the back of the barrel to several inches in front of the muzzle. The receiver accepted modified magazines from the M1911 pistol.

The weapon would eventually see combat with commando units around the world. During World War II, the gun saw service with British commandos disrupting German operations on the continent, and would eventually turn up in the Burma campaign for use against Japanese forces, too. The weapon remained in service with British forces through the Korean war and Malayan Emergency, too. There are vague reports that examples of the weapon may later have popped up in the Vietnam War as well, in the hands of US Special Forces, though as with all clandestine operations, details are thin at best.

Overall, the De Lisle carbine proved the value of focusing a design around a single specific purpose. It was intended to be an ultra-quiet weapon that was designed around the suppressor, to great success. Contrast this with other designs which fitted suppressors after the fact to existing guns, which delivered less impressive results. The gun stood out from its rivals because it was designed differently from the outset, and it proved valuable for this very reason.

Featured image: “De Lisle carbine” by [Atirador]