Bicycle Adds Reliability With Second Chain

February 3, 2025 by Bryan Cockfield 3 Comments

Ignoring the International Cycling Union‘s mostly arbitrary rules for what a bicycle is “supposed” to look like (at least if you want to race), there are actually reasons that the bicycling world has standardized around a few common parts and designs. Especially regarding the drivetrain, almost all bikes use a chain, a freewheel, and a derailleur if there are gears to shift because these parts are cheap, reliable, and easy to repair. But if you’re off grid in a place like Africa, even the most reliable bikes won’t quite cut it. That’s why a group called World Bicycle Relief designed and built the Buffalo bicycle, and the latest adds a second gear with a unique freewheel.

Bicycling YouTuber [Berm Peak] takes us through the design of this bike in his latest video which is also linked below. The original Buffalo bicycle was extremely rugged and durable, with a rear rack designed to carry up to 200 pounds and everything on the bike able to be repaired with little more than an adjustable wrench. The new freewheel adds a second gear to the bike which makes it easier to use it in hilly terrain, but rather than add a complicated and hard-to-repair derailleur the freewheel adds a second chain instead, and the rider can shift between the two gears by pedaling backwards slightly and then re-engaging the pedals.

Of course a few compromises had to be made here. While the new freewheel is nearly as rugged as the old one, it’s slightly more complex. However, they can be changed quite easily with simple tools and are small, affordable, and easy to ship as well. The bike also had to abandon the original coaster brake, but the new rim brakes are a style that are also easy to repair and also meant that the bike got a wheel upgrade as well. Bicycles like these are incredibly important in places where cars are rare or unaffordable, or where large infrastructure needed to support them is unreliable or nonexistent. We’ve seen other examples of bicycles like these being put to work in places like India as well.

Thanks to [Keith] for the tip!

Continue reading “Bicycle Adds Reliability With Second Chain” →

More Details On Why DeepSeek Is A Big Deal

February 3, 2025 by Donald Papp 12 Comments

The DeepSeek large language models (LLM) have been making headlines lately, and for more than one reason. IEEE Spectrum has an article that sums everything up very nicely.

We shared the way DeepSeek made a splash when it came onto the AI scene not long ago, and this is a good opportunity to go into a few more details of why this has been such a big deal.

For one thing, DeepSeek (there’s actually two flavors, -V3 and -R1, more on them in a moment) punches well above its weight. DeepSeek is the product of an innovative development process, and freely available to use or modify. It is also indirectly highlighting the way companies in this space like to label their LLM offerings as “open” or “free”, but stop well short of actually making them open source.

The DeepSeek-V3 LLM was developed in China and reportedly cost less than 6 million USD to train. This was possible thanks to developing DualPipe, a highly optimized and scalable method of training the system despite limitations due to export restrictions on Nvidia hardware. Details are in the technical paper for DeepSeek-V3.

There’s also DeepSeek-R1, a chain-of-thought “reasoning” model which handily provides its thought process enclosed within easily-parsed <think> and </think> pseudo-tags that are included in its responses. A model like this takes an iterative step-by-step approach to formulating responses, and benefits from prompts that provide a clear goal the LLM can aim for. The way DeepSeek-R1 was created was itself novel. Its training started with supervised fine-tuning (SFT) which is a human-led, intensive process as a “cold start” which eventually handed off to a more automated reinforcement learning (RL) process with a rules-based reward system. The result avoided problems that come from relying too much on RL, while minimizing the human effort of SFT. Technical details on the process of training DeepSeek-R1 are here.

DeepSeek-V3 and -R1 are freely available in the sense that one can access the full-powered models online or via an app, or download distilled models for local use on more limited hardware. It is free and open as in accessible, but not open source because not everything needed to replicate the work is actually released. Like with most LLMs, the training data and actual training code used are not available.

What is released and making waves of its own are the technical details of how researchers produced what they did, and that means there are efforts to try to make an actually open source version. Keep an eye out for Open-R1!

Communicating With Satellites Like It’s 1957

February 3, 2025 by Bryan Cockfield 9 Comments

When the first artificial satellite, Sputnik, was put into orbit around Earth, anyone in the path of the satellite could receive the beeps transmitted by the satellite provided they had some simple radio equipment. Of course, there was no two-way communication with this satellite, and it only lasted a few weeks before its batteries died. Here in the future, though, there are many more satellites in orbit and a few are specifically meant for ham radio operators. And, like the ’50s, it doesn’t take too much specialized equipment to communicate with them, although now that communication can be two-way.

The first step in this guide by [W2PAK] is to know where these satellites are in the sky. The simplest way to do that is to use a smartphone app called GoSatWatch and, when configured for a specific location, shows the satellites currently overhead. After that it’s time to break out the radio gear, which can be surprisingly inexpensive. A dual-band handheld is required since satellite uplink and downlink can be on different bands, and the antenna can be made from simple parts as well as [W2PAK] demonstrates in a separate video. Combined, this can easily be done for less than $100. [W2PAK] also goes over the proper format and etiquette for a satellite contact as well, so a new operator can pick it up quickly.

Using satellites as repeaters opens up a lot of capabilities when compared to terrestrial communications. Especially for operators with entry-level licenses who are restricted to mostly VHF and UHF, it adds a challenge as well as significantly increased range compared to ground-based repeaters and line-of-sight communications. There are plenty of activities around satellites that don’t require a license at all, too, like this project which downloads weather imagery from weather satellites.

Continue reading “Communicating With Satellites Like It’s 1957” →

Examining The Vulnerability Of Large Language Models To Data-Poisoning

February 3, 2025 by Maya Posch 8 Comments

Large language models (LLMs) are wholly dependent on the quality of the input data with which these models are trained. While suggestions that people eat rocks are funny to you and me, in the case of LLMs intended to help out medical professionals, any false claims or statements dripping out of such an LLM can have dire consequences, ranging from incorrect diagnoses to much worse. In a recent study published in Nature Medicine by [Daniel Alexander Alber] et al. the ease with which this data poisoning can occur is demonstrated.

According to their findings, only 0.001% of training tokens have to be replaced with medical misinformation to order to create models that are likely to produce medically erroneous statement. Most concerning is that such a corrupted model isn’t readily discovered using standard medical LLM benchmarks. There are filters for erroneous content, but these tend to be limited in scope due to the overhead. Post-training adjustments can be made, as can the addition of RAG, but none of this helps with the confident bull excrement due to corruption.

The mitigation approach that the researchers developed cross-references LLM output against biomedical knowledge graphs, to reduce the LLM mostly for generating natural language. In this approach LLM outputs are matched against the graphs and if LLM ‘facts’ cannot be verified, it’s marked as potential misinformation. In a test with 1,000 random passages detected issues with a claimed effectiveness of 91.9%.

Naturally, this does not guarantee that misinformation does not make it past these knowledge graphs, and largely leaves the original problem with LLMs in place, namely that their outputs can never be fully trusted. This study also makes it abundantly clear how easy it is to corrupt an LLM via the input training data, as well as underlining the broader problem that AI is making mistakes that we don’t expect.

Illustrated Kristina with an IBM Model M keyboard floating between her hands.

Keebin’ With Kristina: The One With The Keyboard Configurator

February 3, 2025 by Kristina Panos 6 Comments

Have you ever wished you could experiment with different layouts super easily, just by adding or removing a few switches here and there and printing a new case? Well, [heyisjambo] says that it’s more than possible with menura, the modular keyboard system.

A collage of menura keyboards, which are modular via the VIK standard. — So many lovely options! Image by [heyisjambo] via GitHub

[heyisjambo] is happy with 36 keys, but is reduced-count-curious and wanted a way to explore without a lot of wasted time and PCBs.

At the same time, [heyisjambo] wanted to experiment with split vs. uni-body construction, and especially the different shapes that are possible when tweaking the angle and distance between them.

And as if that weren’t enough, there’s support for [Sadek Baroudi]’s VIK standard for interfacing data between PCBs, which calls for an FPC 12-pin, 0.5 mm pitch connector and allows for ultra-cool magnetic connectors. This way, you can easily add things like displays, trackpads, and trackballs in the between the halves.

Thanks for the tip, [calculus]!

Continue reading “Keebin’ With Kristina: The One With The Keyboard Configurator” →

Underwater Robotics Hack Chat

February 3, 2025 by Dan Maloney 2 Comments

Join us on Wednesday, February 5 at noon Pacific for the Underwater Robotics Hack Chat with Tony White!

Almost anywhere you look, there’s a good chance you can see a robot at work. Whether they’re sweeping your floors, delivering a snack, building a car, or even driving one, robots are everywhere on this planet. And since over 70% of this planet is covered in water, it makes sense that robots should be there, too. Getting a robot to work underwater at all is one thing, but getting it to work underwater reliably can be quite a challenge. Water always finds a way to ruin your day, after all, and this reality only worsens when you add a little salt into the mix.

Tony White knows the marine engineering field well, having worked in the space for over a decade. He’s currently an applications engineer at Blue Robotics, where he’s worked on everything from full-size autonomous surface vessels to underwater swarm robots. He’s stopping by the Hack Chat to talk about the harsh engineering realities of underwater automation, so if you’ve ever wanted to take the plunge, you’ll want to come to this Hack Chat for sure.

Our Hack Chats are live community events in the Hackaday.io Hack Chat group messaging. This week we’ll be sitting down on Wednesday, February 5 at 12:00 PM Pacific time. If time zones have you tied up, we have a handy time zone converter.

A Cordless Soldering Iron With A Difference

February 3, 2025 by Jenny List 18 Comments

Many decades ago, when soldering was an activity more often associted with copper fabrication than with electronics, a soldering iron would have been a large lump of copper on a shaft, with a wooden handle. You would heat it in a gas flame, and use its pointed end for your soldering. Electric irons have made this a thing of the past, but the basic idea is still one with some merit. [Shake the Future] is here with a modern take on such an iron, one that is heated in the microwave oven.

The business end of the iron is a normal soldering iron bit, but behind it is a piece of sintered silicon carbide, wrapped in ceramic fibre and covered with Kapton tape and a high-teperature-resin 3D printed shield. On the back of that is a 3D printed handle. The whole thing is put in the microwave oven for a few tens of seconds to heat to temperature, and thereafter you have however long the thermal mass of the silicon carbide holds the temperature, in which to do your soldering.

It’s an interesting idea which we can see has some use in situations where you need an iron for a quick job away from your bench but within reach of the kitchen. We like the lateral thinking, and it’s certainly fascinating to see the construction. But in an age of USB-C power packs and irons we have more convenient soldering on the go, so we’re not sure how useful it would be to us.

Silicon carbide is an interesting material, it’s not the first time we’ve written about it being used in a high temperature application.

Continue reading “A Cordless Soldering Iron With A Difference” →