Lots of progress photos for Dexter Starfighter’s ED-E build. The body/head is essentially a HDPE milk jug, 3D printed parts and foam mat pieces glued with contact cement, portions covered with craft foam and seams covered with air drying foam clay, primed, and painted. There are also some 3D printed connectors that go inside the “head” and at the “neck” to keep everything mounted and steady. We used the milk jug because it was the right size, approximately the right shape, the bottom of the jug already had a “grill” like look similar to the ED-E robot face, and it was free – since it would have been recycling/garbage otherwise. In order to assist with reinforcing the body1 , we cut a second jug and inserted most of it into the outside body. One downside to the jug is that HDPE is slick and doesn’t really take paint, glue, or even contact cement particularly well. Over Maker Faire weekend the paint started to fleck off in places. I think the size and shape is great – but that perhaps we would be better off covering it in paper mache and then painting and gluing onto that.
The old ED-E body was roughly the size of a tennis ball2 and the new one significantly roomier for the electronics. The PVC pipe had some holes drilled into it and was held at the right height using a zip tie. It held up really well all weekend and was apparently more comfortable than a shoulder-mounted robot, but it was a fair bit more to carry around. We’re working on some wheel based designs – and I even bought an IV stand off Craigslist for $10 to potentially use as a base for next year. 3
This slideshow requires JavaScript.
This year we opted to really focus on the “cosplay” elements of our project, making ED-E much larger and designing custom PIP-Boys. The PIP-Boy isn’t screen accurate, has an olive-green body which is more reminiscent of the early video games, and is still missing several of the greebles… but I really couldn’t be happier with how it came together and turned out. We were easily recognizable as Fallout vault dwellers everywhere we went.
While I did design this model entirely from scratch, I did take inspiration from various photographs from the show, game, and other 3D models out there. In particular, I liked the models from YTec3D, the official Bethesda props, and some of the features I noticed from various Etsy sellers.
I’d designed the PIP-Boy case to have a lot of interior space and specific cutouts that should work well for adding LED’s and real buttons and knobs later. I learned a lot from the Raptor FX guys about building with craft foam and foaming TPU, so I think the next iteration will include 3D printed cuffs and/or a buckle-based attachment system.
Five versions printed for testing and fitting, a failed orange PETG print, and a final orange PETG print
Final OpenSCAD render, top down view
Final OpenSCAD render, oriented for printing
I couldn’t be happier with how this model and print turned out.
Fifth variation on the 2.5″ SATA hard drive caddy for the Lenovo ThinkCentre Tiny M70q case
The part pops into place snugly, even though it doesn’t include any snap-fit parts. The two tabs on the right fit over and around two bent metal tabs coming off of the frame, the beveled section on the left sits on a metal bar, with a hole for a bolt head which holds that metal frame in place. The design doesn’t include all the fancy fiddly bits that go in and around that metal bar, but it doesn’t wiggle with medium pressure. It’s easy enough to drop into place and remove without too much trouble.
In a perfect world, I would have printed it in a more appropriate color befitting an arc reactor, like glow in the dark, white, light blue, or some combination. However, while I have a medium sized collection of PLA in various colors, I’m fine with just using this one roll of PETG in it’s bright orange for those use cases requiring something that requires more flexibility or heat resistance than PLA.
I’m super happy to have this PC housing a drive that will likely last me a very long time. Our old external backup drive holds 2TB, is quite slow, requires a USB mini to USB A cable, and a huge wall wart power adapter. Now I can keep our files easily accessible inside this PC without no fussing with cables/adapters, taking up a USB port, and not requiring the additional space that’s literally larger than this entire PC.
In other news on this “Project Jarvis”… I haven’t been making much progress lately. I’m not able to run any reasonably capable local LLM’s at any exciting speeds, but it’s still at least possible at a 1-2 token per second rate. I think once it’s set up I’ll probably want to switch it to a “dumber” smart assistant kind of system where it helps me answer basic verbal queries about the weather, date, schedules/cron jobs, and any sensors I might create.
I started to write this blog post a few months ago, and just published a post pontificating / complaining about AI search result slop.
Google has become nearly useless lately. Searching for anything turns up it’s AI slop nonsense or advertisements rather than actual links to something useful. Google’s initial promise – high quality uncluttered search results is dead. Any results are now hidden below their AI slop, advertisements, and other results from other slop sources. I might as well publish it now, rather than have it relegated to a draft for years to come.
If it were possible for Google AI summaries of search results to be infallible, I guess that would be fine – but it’s just so far from the truth. I’ve yet to see an LLM that doesn’t just try to generate something to please – rather than try to generate something that is accurate. Google AI offering false but pleasing slop reminds me of the Azimov story Liar!. Without a “three laws of robotics” style filter, there’s no systemic guardrail preventing automated misinformation.
While I think there’s room for legitimate practical LLM uses, shoving it into search results seems like a complete waste for everyone concerned. When people say “ChatGPT/Claude told me X”, that’s really no different than saying “ChatGPT/Claude thought I would be pleased if it said X.” But using an LLM1 for synthesis and summarization of information and data seems only able to generate either results its hopes are pleasing to the requestor or results hopelessly biased by it’s training. When too many people have the wrong idea about what an LLM is, how it works, or how to interpret it’s responses, putting an AI result before actually informative links is really damaging.
Yes, I’m still struggling with the implications and ethics of LLM use.
Small LLM’s you can run at home or offline, on your machine, in your browser, or on your phone feel intuitively less evil. But, they’re just small distillation spawn from a much larger LLM – built from hoovering an internet of data, all social media sites, literally destroying old books for even more data, building hundreds of data centers, consuming power and water, churning out waste and noise. If I can’t justify using a big AI company services, shouldn’t I also refrain from using their derivatives as well?
I am still using frontier AI, just not paying for it in the normal sense of sending them money. But, I’m under no illusions that any time I’m using their services, I’m helping shovel content into their machines and train SkyNet. Since small models can be run entirely offline, they don’t have the same “you’re the content” problems.
One of my goals is to bootstrap my way out of using these things. I’m still very much a Linux newb and still working on my Python skills and what little programming abilities I have are all self-taught. StackOverflow used to be a reasonably reliable source for finding discussions on how to overcome a given technical problem, but now it’s a ghost town. If I can’t turn to something like StackOverflow… the options these days are asking on Mastodon and hoping a mutual replies, asking / searching on Reddit and hoping the answer isn’t a ‘bot, or just asking a ‘bot directly.
It’s still the same problem – and I honestly can’t tell if I’m trying to rationalize my LLM use or if there are any legitimate uses. Certainly OCR and interpreting MRI and other radiological reports to help save lives are reasonable and beneficial uses. I guess part of my struggle is that I literally cannot avoid interacting with stupid AI when they’re everywhere, in everything, ads in media and billboards, spam emails/texts, website chatbots, preliminary phone / chat bot screenings, and even shoved into the products I’ve owned for a while.1 If I’m already forced to interact with an LLM to (perhaps) benefit other people / companies, why shouldn’t I feel comfortable using them to (a) filter out their AI garbage (b) assist me with minor tasks?
Possibly unrelated, both of my kiddos are facing AI nonsense in school. The eldest receives constant warnings at university about what uses are permitted / not permitted – and then the university has students submit their work through an AI port to determine whether the students are using AI. 2 The younger one has teachers saying their work will be graded by AI – in junior high?!
I refuse to believe these students’ work aren’t being used to train the next generation of AI education software.
I know I’m rambling. I recently sent a picture of a basket of baby otters to some family members – who thought it was probably AI. I had seen it on Mastodon and based on the context, I felt it was likely real – but I had to question it. Performing an exact match image search didn’t work – since the same image has been spammed across the internet in the AI era of post-2023. I narrowed the search to only those results prior to 2020 and was able to locate a copy of the image from roughly 11 years ago.34 This made me think – since current Google searches have become complete AI slop / summary garbage, why not narrow all internet searches to only those results prior to 2023?
I think where I’m leaning on this entire ethical morass is … this stupid AI nonsense is everywhere around me, my own work and thoughts are going to be fed into the giant provider AI machines willingly5 , begrudgingly6 , or unwillingly,7 and I don’t have a big problem using similar technology to build systems to filter out AI garbage and teach me further skills.
Technically, anyone can record in public and by going out in public I’m consenting to the possibility my actions and likeness will be fed into the giant data machine. Even so, still Flock + Palantir + Meta/Google/etc is about as offensive as big brother / big data gets [↩]
Assuming all human knowledge and the entire of the internet is fed into LLM’s, they’ve digested the thousands of blog and social media posts I’ve created. robots.txt. What a joke. [↩]
This little banjo ditty from Sesame Street circa 1979 permanently lives in my brain. I’ve played the video for my kids and it’s even become something of a “meme” in our household when someone needs more room.
Anyhow… I want to add more storage space to my Lenovo ThinkCentre Tiny M70q computer. The problem, as one might infer from the name of this particular form factor PC, is that it is pretty tiny and there’s limited space inside. Even so, I have to say it is a very well designed PC case. The PC cover comes off after loosening a single screw in the back1 , hard drive caddy clips into place, the cover for where the RAM sticks are located can also be removed by hand without any tools. It’s just so incredibly modular that swapping out either RAM or a drive would easily take less than a minute. These particular kinds of PC’s appear to be marketed towards office and corporate systems, where there’s an IT department with a box of parts, that can just quickly disconnect a PC, drop in a replacement, or apply a quick fix. With their quiet operation, sturdy metal frame, stock features2, modularity, and price point make them a perfect choice for any office environment. I’m guessing the reason these haven’t taken over the consumer market is because they’re not sexy and don’t have any flashy LED’s.
It also has two small features which I love. The PC comes with a small integrated speaker – so you can play a small sound or sound file without any peripherals. 3 Also, you can buy these very inexpensive little expansion boards to give your PC an extra HDMI, DP, or COM port. I know this might sound like I’m shilling for Lenovo, but I picked this thing up on the secondary market for a hair over $200.
The problem I’m running into is that the stock PC hard drive caddy can only accommodate drives up to about 7-10mm thick and the most storage space you can get in a 7mm drive is 2TB and anything above that pretty much necessitates a 15mm thick drive.
From just a visual inspection456 I believe I can physically insert a 15mm thick drive into the SATA drive area. A quick search of Reddit suggests the only way to do this is to be a complete maniac. One Redditor said he sanded out the tabs which would normally hold a 7mm drive in place, flipped the drive upside down to pop into the metal pins, then had to twist the SATA cable around to connect to the upside down drive. The drive caddy I have doesn’t have those tabs, so it should be possible to mount a thicker drive upside down and put a half twist7 into the ribbon cable, but… I’m not excited about twisting that thin delicate ribbon cable. They’re not super expensive, but they are proprietary and the hassle of putting the device out of commission for a few days while a replacement arrives is not something I feel like risking.
I started searching ebay to see if there were any off the shelf parts to drop a 15mm drive into a ThinkCentre Tiny without success. I did, however, discover this absolute gem for sale on ebay for $20 plus $10 shipping:
Lenovo ThinkCentre Tiny 2.5″ SSD Tray/Caddy by Ropsu11 on MakerWorld
I express no opinion on whether $30 is a good deal for this product, but I couldn’t help notice it certainly has the look of a 3D printed part, which lead me to what I believe is the original file by Ropsu11 on MakerWorld. This looks perfect for my purposes, except that the design would still need to be altered slightly so that I could make it work with a 15mm drive without having to flip it on it’s back.8 I tracked down the specifications for 2.5″ SATA drives so I could figure out just how to adjust the 3D model file. Apparently the screws for SATA drives, irrespective of whether it’s a 5mm or 15mm tall drive, are located 3mm from the bottom of the drive.
2.5″ Form Factor Drive Dimensions (SFF-8201, established 01/19/2018)
The 3D model file was available on MakerWorld as an 3MF file, which I rotated slightly and then converted into an STL using PrusaSlicer. By using a little bit of OpenSCAD magic, I can take the model to center it on the origin. In doing so I discovered the various holes/tabs don’t really seem to line up properly. In an ideal world, I’d be able to adjust the original design files and failing that email or message the creator. Since the creator doesn’t enable messages from users with less than a level 7 on MakerWorld, it looks like I’ll need to try and either adjust or redesign the file myself. The specific locations of the mounting holes should be determined by the SATA drive specifications above – rather than slightly offset by various small amount. By taking the model and mirroring it over the X and Y axes and using the “%” background modifier, I can make sure it’s perfectly centered – but ended up discovering the various holes don’t seem to really line up with one another.
After some fiddling with the STL for that file, I find myself where I usually do after inspecting someone else’s model. I think I’ll just design it myself. Yes, there will be some measurements, possible repeat printing, but I’ll be certain it will fit my machine and new drive – and I can customize it as well.
The only things I have plugged into this device are the stubby wifi antenna, the dummy HDMI plug, the power cable, and sometimes a USB flash drive or device I’m working on. I just want this PC to exist invisibly and silently in the back of my home office doing it’s thing. [↩]
I had one incredibly pompous “doctor” use the word “oculometrics” to describe how he was able to perform a medical examination and make particular measurements. In looking up the word now, it appears it refers to the measurements of eye movements – not one’s ability to measure with just their eyeball perception aided by their brains. [↩]
That, of course, would be considered “cerebral aided oculometrics” 🤮 [↩]
I say “doctor” not because this person was a time traveler with dubious credentials, but because they held themselves out as a doctor of chiropractic. I’m sure people find relief through chiropractic manipulations, but as an orthopedic surgeon once told me … “90% of all back pain goes away on it’s own. So, just because you rub bleu cheese on your back and pain goes away, doesn’t mean it worked.” [↩]
One of my favorite things to do leading up to Maker Faire is to reach out to interesting people / projects / exhibits ahead of time to learn more about what they’re doing, how they make things, and be sociable, really. There’s so much to see that even if your Maker Faire enjoyment strategy is to sprint from booth to booth all day long talking to as many people as possible, it’s very easy to miss someone who was only there for just one day or part of the weekend or even someone who’s tucked away in some small corner or behind a booth that gets swamped.1 I reached out to RiseLink about their incredibly cute looking board and workshop and they graciously offered to send me one of these boards for experimentation – with their only request being that I share a short video of whatever I build.
RiseLink.AI’s AI Talking Toy
RiseLink doesn’t seem to be a “maker-facing” business per se, but rather a big chip maker creating some of these development boards to showcase their chips. From what I can tell, this board packs a huge punch for the $50 price tag – a built in camera, microphone, speaker, LCD display, and wifi capabilities, fully pre-assembled into a very cute little package.
Suffice it to say, I’m having a lot of problems getting this project off the ground. Here’s how I approached this project:
Locate Documentation
The documentation is somewhat sprinkled throughout the RiseLink website and their Github. For the moment, I’m just going to dump all the links and sort them out later.
cd /tmp
wget https://developer.arm.com/-/media/Files/downloads/gnu-rm/10.3-2021.10/gcc-arm-none-eabi-10.3-2021.10-x86_64-linux.tar.bz2
sudo tar -jxvf gcc-arm-none-eabi-10.3-2021.10-x86_64-linux.tar.bz2 -C /opt/
Clone Core SDK Repo
mkdir -p ~/armino && cd ~/armino
git clone https://github.com/bekencorp/bk_avdk_smp.git
Compile Base Audio/Voice Firmware
cd ~/armino/bk_avdk_smp
make bk7258 PROJECT=doorbell_asr
Verify Output
ls -lh build/bk7258/doorbell_asr/package/all-app.bin
Flashing the Board
Find the USB port (likely “/dev/ttyUSB0”)
ls -l /dev/ttyUSB* /dev/ttyACM* 2>/dev/null
Ensure user permissions for serial access
sudo usermod -aG dialout $USER
Flash firmware
cd ~/armino/bk_avdk_smp
tools/build_tools/armino -p /dev/ttyUSB0 burn -f build/bk7258/doorbell_asr/package/all-app.bin
By the end of all this2 , I can receive “success” messages when I flash the firmware, but I can’t get the device to do anything useful yet. I’d really like to get this working and will keep at it. Wish me luck.
… Uh, as I was getting ready to publish, I figured I’d add some visual interest to this post and looked at the RiseLink.AI’s website and blog. I found they linked to a Google Doc with a tutorial for getting it working. When I approach this again with fresh eyes, I guess I’ll start there!
As time grew close to Maker Faire, we had to sharpen / tighten our goals. We always try to improve some part of our projects and change things up. This year, we worked on the more “cosplay” elements and left the more electronic and technical aspects largely alone.
ED-E Companion Robot
Dexter built an entirely new body for ED-E from a milk jug, 3D printed parts, foam mats, and model magic. In retrospect, an HDPE milk jug is so smooth and adhesive resistant that it may not have been the best choice. The paint flaked off in some small places and the 3D printed internal structures didn’t stay in place as desired, but everything held together quite well over the course of the entire weekend.
ED-E used to be built into a small 3D printed enclosure, about the size of a baseball, which was limiting for power supply and other mechanical abilities. This year we mounted him on a PVC pipe, but quickly learned the end needed to be padded otherwise the constant shock of hitting the ground would slowly scramble his brains.
I think an entirely 3D printed case is very doable for next year, would take pain much better, permit easier access to his brain, better placement for LED’s and a speaker. He would probably benefit from using separate ESP32’s for the brain and the PIP-Boy to communicate wirelessly, so that the robot could be easily untethered from the arm.
Another issue with the robot was simply carrying the thing over the day. The robot and PVC aren’t that heavy, but over time lugging it around, even using it as a walking stick takes its toll. The temporary fix at Maker Faire was to tape a bunch of foam to the bottom of the PVC pipe. One possible fix would be to design a TPU cap to go over the bottom of the pipe, to cushion the impact of the pipe on the ground – and help reduce the impact on the robot components. One idea is to add a wheel or caster to the base and another idea would be to put the whole robot on an IV pole. Putting the pole on a single caster/wheel could be problematic if we needed to set the robot down – unless the wheel had a retractable kickstand or retractable wheel. An IV pole might work well, but sourcing one might be an issue. 1
Vault Dweller
I pivoted my project from being solely about a shoulder mounted robot to being more of a “vault dweller with companion robot.” I designed our PIP-Boys from the ground up, making several changes from other designs out there for aesthetic and practical reasons, while keeping the essential look. Although I didn’t get around to adding yellow accents to my blue jumpsuit, I was recognized throughout Maker Faire and called out as a “vault dweller” probably due to the icon combination of the jumpsuit and PIP-Boy.
My PIP-Boy included the e-ink display I had originally purchased for use in a kind of digital to-do list, but disconnected from a microcontroller. I’m looking forward to building in additional features to the device, adding a power supply, microcontroller, buttons, and lights to bring it to life.
I normally take a sling backpack with me when I’m out for a day trip or running some errands. However, after the first two days I skipped it entirely since my jumpsuit had ample pockets and storage. The various vault suits out there don’t seem to include pockets, but that’s just silly. Ideally, for aesthetics the vault suit would feature some leather / faux leather (possibly weathered craft foam) pads at the elbows and knees as well as a brown / dark brown leather belt with pockets / pouches for stuff.
I’d really like to get a working set of buttons and microcontroller working in my PIP-Boy to be able to change the display. There’s probably room in the PIP-Boy for a battery pack, some wires going to various LED’s2 , possibly a small servo, potentially a little laser3 , and maybe a little spring loaded projectile launcher.
I couldn’t be happier with how the PIP-Boys turned out. My 3D designs really came together. The cuff parts were just craft foam / floor mats and fit reasonably well. They had a tendency to slip or rotate on the arms a little, but not to a problematic extent. The 3D printed parts stayed affixed to the painted foam cuffs well enough. Both were instantly recognizable as PIP-Boys by lots of people. I think with some sanding, wet sanding, and a blast from a heat gun, they’d look even better. I’m really looking forward to sprucing these up.
I don’t mind paying ~$10 for a used one, but not interested in picking up a brand new one for $40-400. I did find one on Craigslist, but the seller hasn’t replied after several hours. After this afternoon, I’ll be a lot less immediately interested since we probably won’t be suiting up again for another year. [↩]
I have a great spot for the 5mm NeoPixels I bought years ago [↩]
I’ve broken this year’s Maker Faire roundup into two sections – those things that were totally new to me – and those projects and friends I was fortunate enough to see again.
Fred Zyda’s Harmonograph
Fred and Eleanor made a harmonograph to draw Lissajous figures based on designs by Karl Sims. Karl also published a tutorial on Make Magazine a few years back. This is one of those projects which doesn’t look like much on paper – but when it’s in front of you is downright mesmerizing. One of the things I love about Maker Faire is meeting people who have exhaustively searched for the perfect parts for their project. Here, Fred bought up pretty much every kind of pen possible before settling on the Pentel Energel Pro Permanent Gel Pen 0.7 mm – the key being not just a smooth motion, but a pen that could draw swiftly enough to keep up with the machine.
Pentel EnerGel Pen Refill 0.7 Metal Tip (#LR7-A)
Maclean’s super accurate Apollo space suit
Maclean‘s Apollo spacesuit was a marvel of hyper accurate multi-mode making using machined parts, FDM and resin 3D printing, vacuum forming, sewing, and electronics. he also brought his Ghostbusters proton pack!
El Pulpo Magnifico
The photographs and videos of this metal monstrosity simply do not do it justice. Their DJ spun some crazy tunes on Saturday night and their FJ1 were blowing the lid off the evening with huge bursts of flames while Makers ate, partied, and danced all around under a moonlit night.
The booth for Generate Labs is what making and helping people is all about. Makers and medical professionals work together to come up with ways to quickly and cost effectively solve real world problems and help out individual patients in ways that might otherwise be completely impossible. I was touched by their stories of helping a child with cerebral palsy play his favorite Taylor Swift songs, someone else play their favorite video games by building a prosthetic that would otherwise have been $70k, building life size physical representations of the human circulatory system so medical professionals could visualize how implantations are placed, special prosthetics so that a baby born with differences could learn to crawl. Even though their facility gives these devices away for free, I have no doubt they’re actually saving the facility tens of thousands of dollars by being able to print color coded test tube trays and repair their microscope equipment.2
These two video games were both clearly passion projects. Future gaming consoles may never have teat or foot interfaces, but these were both incredibly fun games and super enthusiastic teams.
This group had a laptop connected to an LLM, which pulled an image, which was turned into a simple vector drawing, then printed out using a 1980’s plotter. While LLM use tastes like the opposite of making, it was less a critical component of their project and more about enabling a kind of pictionary telephone-game between booth visitors.
This university’s pitch was quite compelling – a four year university that is less expensive for out-of-state residents than it is for a local California community college, with no less than four separate makerspaces.
One of my favorite things to do at Maker Faire is basically sprint from booth to booth, learning everything I can about what people are doing and how they overcame challenges. Adrian took great care with his Lumotron to orient the 3D pieces to minimize any tell-tale appearance of 3D printing artifacts – and used some interesting mathematic applications4 to run a single sided LED strip through the entire sculpture in such a path that it created essentially a double-sided strip throughout. Plus, he built his own javascript tools for planning it all. Sometimes I like having a thing… but so much more often, I just feels so much more satisfying to understand how a thing came to be.
I reached out to Ravi before Maker Faire because I was so interested in his project. If you’re not using a laptop to watch videos, there’s no real reason for a big power-hungry screen when a simple low power e-ink display could help a device last for days or weeks. He also hooked us up with a some foam for ED-E’s PVC pipe! Thanks bud!
Chris spoke passionately about building a life-size 50 foot tall Iron Giant out of steel. It took some serious love and determination just to do the head – and I have no doubt he’ll be able to build the entire rest of the robot.
These were really nice looking metal EDC utility knives; thin, a smooth bistable movement, and it made use of easily disposable and replaceable utility blades. If I was a knife carrying kinda guy, I’d probably want one.
I like to peruse the Maker Faire website before the event starts, scoping out all the things I’d be interested in seeing. Dipstick caught my eye because it was a fun solution to a very niche issue (wanting to see clearly into shallow water) with some fun pictures. It’s basically a flexible digital underwater periscope and was an interesting combination of electronics, 3D printing, and plumbing. The father and daughter duo were enthusiastic and engaging as they explained how they made it. They purchased pretty much every kind of digital endoscope before finding – and then having built – the exact kind of waterproof one they needed for the project!
Warren creates block prints using 3D printed versions of his digital designs – but also drawings / paintings over trading cards and pages from old books. Most years I don’t buy anything at Maker Faire besides food and drink – but this year I bought one of Warren’s pieces because every time I looked at it I just laughed.
The booth was so mobbed that I only got a chance to watch it in action. The project was a cross between a carnival game and Simon. You and three competitors slap one of the four colored buttons matching the colored lights as they flash beneath your balloon.
When I first saw this page on the Maker Faire website, I thought “So what? A single servo in a ball wiggling?” When I saw it in real life… I realized Fangwei was able to actually control the approximate direction of the sphere just through wiggling that one motor! In a delicate balance of the placement of the electronics and felt feet mixed with shaking the servo in a particular way to one side or the other and some careful calibration – he was able to actually move the robot in a desired location.
I didn’t get a chance to actually build one of these, but I watched the videos on their Maker Faire page – and got to see several in action. They’re so simple, so fun to watch, and open themselves to lots of design and exploration by adjusting the paper fins.
This is another exhibit that almost doesn’t make sense until you see it. Kids make drawings on transparencies, these are folded and joined into a ring of four panels, then placed on a rotating mechanism with a light source and then behind a diffuser. The result is a dreamy kind of pantomime repeating visual story.
Brain at SnackNotes had a LOT of different little projects to show off in their small booth. He had zoetropes, a DIY whiteboard etch-a-sketch, squishy blob puppets, and lots more – and was happy to share everything about their projects.
We also got to see some friends and favorite familiar projects from prior years:
I don’t think I’ve seen Paul Nosa at Maker Faire for some time – but it was great to see his sewing artistry on display.
Melty Cube. Daniel’s “melty cube” has to be seen in person to be believed. As the staff spins and changes colors, you realize each face is slightly different, causing them to become animated before your very eyes.
Clever Clocks. I saw these clocks for the first time in 2025 and was mesmerized by their seeming simplicity and abstract beauty. Every single one of their clocks is clearly a work of art. What I love about these timepieces is that when you don’t need to know the time, every one of them is lovely kinetic sculpture.
The balloon and buoy projects are not at all affiliated with one another – but I love how they’re both ways that regular people could accessibly explore the entire planet.
We’ve bought several pieces from the Dickinsons over the years and love checking in with them each time we come to Maker Faire. His work is beautiful and it’s mesmerizing to watch him craft something out of a blob of glass and flame.
As with so many makers, everyone at the entire booth is absolutely overjoyed about sharing what they made, how they made it, how they overcame challenges – plus they have some hilarious Hollywood stories as well.
I first saw Stephen‘s word games at last year’s Maker Faire. This description won’t do it justice, but it’s kind of like a combination of Banangrams/Scrabble and Tetris, where you build words using big physical blocks and LED displays, while new letters drop down a screen.
I’d seen Sync Tank in years prior, but didn’t get a chance to stop by their booth again. However, since I haven’t typed up my notes from prior years, it seemed appropriate to add it in now. Chase built a way to watch and monitor his fish underwater. If you love fish and fishtanks, but are bummed you can’t have them with you all the time, this is the project for you.
I last checked out the laser maze about two years prior and didn’t get a chance to go back this year. If you ever get a chance, it’s incredibly fun to try and make your way past these lasers in time to beat the clock.
There were more than one booth from other makers that were largely indistinguishable from AI image / AI 3D model + printing / AI vision / AI API calls [↩]
Hermes Agent and LangGraph probably aren’t going to work for me either. On a highly resource constrained system like my no-GPU/no-VRAM Ubuntu box, Hermes quickly has a problem with long contexts. The benefits of Hermes is that it tries to iterate quickly using lots of calls, but with each call including a huge amount of context overhead about the user, the tools, and the current task, the context bloat means each of these calls takes forever. I like the UI and as long as I’m using an API based LLM through OpenRouter1 , it’s fine. While a small local model can move relatively “fast”, it gets bogged down with context and larger models which can handle the context are even slower. Again, there’s a lot of great things about Hermes with an API. It can be conversational, it’s fairly coherent across long conversations, it can review files, make changes, and does indeed to try to improve itself as advertised. It’s also got a lot of ways to quickly flip between models and sessions/conversations on different topics. Even with all these abilities, it’s still only a partial solution to what I want. I’d like to have something that can run custom scripts on a regular basis, interact via voice, and these features don’t seem well implemented quite yet. 2 If left alone to do a task, it will litter it’s home directory with a pile of different folders and files, despite explicit instructions otherwise. It’s very capable – but can get confused and messy. Perhaps this wouldn’t be a problem if it were powered by a frontier model, but I’m not going to do that.
I tried out LangGraph as a possible replacement, but it has it’s own problems. It’s far more “bare bones”, with all commands needing to be typed into the command line. Adding skills via specific scripts works very well, powering it via local LLM works very well, it’s able to accurately interpret language and make tool calls, properly set up CRUD lists and reminders, and even manages separate conversations in it’s own clunky way. However, it seems to really lack the iterative ability of Hermes to actually get things done. If it had a better UI, a more clear path to having it semi-autonomously work on a task, I think I could make a go of it.
Even though these two systems aren’t for me, I’ve learned a lot about what I like and what I could put to use in the future.
I keep circling back to … maybe I should just build my own thing. I’ve built the various parts several times before getting distracted by something shiny, so it’s really a matter of collecting those pieces, cleaning them up, and wiring most of them together with what I’ve learned from using other systems.
While I’m rambling, I just read this article on Hackaday about the LLM moats for frontier labs starting to shrink. 3 It kind of makes sense these damn frontier AI companies had to buy up all the RAM. If people could easily pick up 128GB of RAM for $2504 and given the performance of medium sized models in the 30-70B / 30-70GB range, no one would pay $20 per month for access to an AI. At the rate models are improving and being released, it certainly seems like the only thing keeping big AI companies afloat is high RAM prices. While I would not spend $3k-$5k to get a 128GB RAM beast today, since it would never pay for itself, if that price dropped to $500, the cost / benefit analysis starts to make sense.
It does make me wonder, though… what happens if RAM supply suddenly increases, home inference machines are suddenly feasible and every domain name / hosting provider has a $5 upsell for their own Qwen 3.8+ instance. What happens to these billion dollar companies? I would think there’s only so much “intelligence” people are going to be willing to pay for. If all people really want are digital receptionists, developers, lead generation, content generation, investment predictions, and the occasional therapist/pseudo-relationship, why would they pay for an astrophysics PhD level hyper intelligent machine?
In a move that might surprise no one – these last two points suggest a much darker path for frontier AI companies. Perhaps the game was always about about increasing surveillance and diminishing critical reasoning of the masses.
I’d been looking at several LLM powered “agent” frameworks recently. OpenClaw got a lot of buzz several months ago, spiked the cost of Mac Mini’s, and it seems reality doesn’t quite live up to the hype. Peter Steinberger’s description of his OpenClaw1 creation in multiple videos, podcasts, and posts was he would wake up each day to discovery the agent had been working tirelessly throughout the night on projects, building things, discover things, bring him new opportunities, and learning and growing as it went. Reddit and Youtube exploded with people snatching up Mac Mini’s for their high unified RAM / small form factor / modest cost, implementing their own OpenClaw’s, and telling everyone about their successes breathlessly.
I think my favorite “review” was this:
The next ClawdBot rebrand?
In the span of a few weeks ClawdBot went through several name changes, Anthropic blocked it from consumer level plans, Steinberger got a job with OpenAI, while people started to post their runaway token costs and concerns the agent was escaping containment, grabbing personal information, publishing personal information, and just generally running amok.
I have never tried OpenClaw because it seemed like an AI vibe coded slop privacy / security nightmare with a thick frosting of marketing slathered all over. I understand the appeal of OpenClaw. It would be so nice to have a little digital butler, looking at my cluttered desktop, helping tidy up, organize me, make some connections and surprising me in the morning with some new developments. After all, where would Tony Stark be without Pepper Potts and Jarvis to keep him on the rails?2 I also understand how some semi-technical people would be dazzled by the promise of something that would work around the clock to make them money, all for the cost of a Mac Mini.
A short interlude about Chik-Fil-A. I don’t like their politics and will not support them. I’ve never once given them a single penny. But, I have eaten food from Chik-Fil-A… When a new located opened not far from my former workplace, they came by and left a HUGE platter of delicious chicken nuggets and sauces. They have historically donated to our local Maker Faire, providing lunch for the makers. Hide their branding and of course their food is going to be delicious – that’s the entire point of fast food. The first year they provided lunch, I didn’t eat it – because I don’t like Chik-Fil-A. The second year… I was weaker – but also more practical? I was hungry, the kids were hungry, and although I had no problem paying for food at one of the many vendors, we’d been there a while, it was hot, the lines were long, and well, the food was right there. I suppose the most subversive thing to do would be to take their food and toss it in the trash – or maybe shuttle their food to the kinds of people they despise. But, I’m not in the habit of wasting or delivering food. My own subversive little act was to eat the food – and resolve to never, ever, advertise for them or give them any money.
Back to LLM’s. I have tried Hermes Agent, running in a docker container in a machine with very little personal information about me. I’ve only been using it for a few days – and I think it has some real promise with some real caveats.
The first caveat is that I won’t use any frontier LLM services because I can’t justify giving OpenAI/Anthropic/Google/Meta money for such things. Whatever Google once was, whatever their motto might have been, they’re so far removed from that as to be a different animal entirely. 3 The obvious downside is that without access to a very fast, very big, and very powerful LLM, my Hermes Agent will absolutely be less useful than someone else’s instance.
It will run in a Docker container on an isolated, budget machine, or not at all.
In practical terms, this means I’ve limited myself to either smaller and less capable local LLM’s that my machine can run or free tier API’s for various LLM providers. I’m still fumbling my way around Linux / Ubuntu and already using LLM’s to get my bearings, so these resources for providing the LLM reasoning power for Hermes feels within bounds too.
As always, I simply don’t trust any LLM’s outputs. I don’t trust their purported facts, claims, summaries, etc. Even so, there’s no arguing they are capable of building systems. Some family recently “fact checked” me using ChatGPT and I pointed out its merely a machine designed to provide them with the very next token of what they want to hear – not a system for giving them facts. I feel like it’s probably impossible to build an AI / AGI without training it on facts, but relying on a token generator for facts seems foolhardy. That said, I don’t have a problem relying on a system / set of processes / code base that I can inspect and verify works to arrive at a solution.
These “agent” systems are usually better called “agent harnesses,” since their best feature is offloading the work of monitoring, guiding, and nudging an LLM to perform a task. The Hermes agent adds a lot of extra features – keeping things organized, keeping certain chats available to revisit later on, switching LLM’s between local / API based, but at it’s core it is saving me the time of just babysitting an LLM.
As understand these agents, the first thing they do is take a query, ask the LLM to break it into pieces, and then get to work on discrete elements. It used to be that on long tasks ChatGPT would keep popping up a “Continue” button. About two years ago I wrote a simple Chrome extension that was active on ChatGPT and would simply auto-click the “Continue” button whenever it appeared. I think this is less a thing these days and ChatGPT will either truncate a reply or simply stop. But, keeping the LLM on track for each small sub-task and spurring it to complete appears to be part of what the agent is doing under the hood.
I’ve tried several LLM’s to assist with writing code, with varying degrees of success. The best OpenRouter free model I’ve used recently has been Poolside’s Laguna S2.1 with 118B parameters, a “mixture of experts” model with only 8B active at any given time. I’ve been very impressed with it’s coherence on long tasks and overall quality of work – and it has been extremely helpful in setting up Hermes. I haven’t tried a recent Qwen or Deepseek model on my very much underpowered system, but I plan on doing so soon.
My first goal with this system is to get it to just run a bunch of python scrips on a cron job. I may need to build up a separate system entirely that only runs cron jobs, but which can have it’s roster of tasks manipulated by Hermes. Too often the LLM gets involved and wants to interpret a query rather than run a static script, which just leads to more hassles. Perhaps a frontier LLM wouldn’t have this problem – but I don’t plan on using them.
NGL, I really hated Gwenth Paltrow’s Pepper Potts and would have liked nearly anyone else in that role. Maybe it was the direction – or maybe just my dislike for Paltrow – but I found her bumbling, flustered character exhausting. The comic book Potts was a capable, intelligent person in their own right. Maybe there’s hope for a better person in a future reboot? Hell, if they can recast Roddy, they can recast Potts. [↩]
I won’t even bore you with the full story of how Google promised a remediation for a Pixel 6 battery problem, bait and switched, then slowplayed it until everything expired. [↩]
