site banner
ArjinFerman  Tinfoil Gigachad  2mo ago (text post)   Edited 2mo ago   760 thread views

Tinker Tuesday for April 22, 2025

This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.

Post your project, your progress from last week, and what you hope to accomplish this week.

If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service

1
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Anyone here an expert on getting image AIs to do things? There are some detailed line chart images that I would like to extract the data from - has the software improved enough to do this yet? Surely, if we can read images well enough to generate cartoon versions of them, we can figure out the numbers on a line chart that are already displayed on a pixelated grid!

I'm a total neophyte in this area, so if it would be possible can you explain like I am truly stupid?

I've had reasonable success with Claude. I've a pro subscription. If you send me the graph or a link I'll give it a go.

That would be amazing. Could you try with the "Jobs in 6 Months" (just above the "(6)") in this link? If you click on the chart, you get a decent-sized image.

https://www.yardeniquicktakes.com/in-10-charts-consumer-confidence-jobs/

Your results are going to depend on two things: What your detailed line chart images look like (including how detailed they are and the resolution) and what ai you use to analyze the images.

I quickly googled "detailed line chart images" and picked a line chart from the results (the second one on this page: https://www.investopedia.com/terms/l/linechart.asp ) and fed the image to both Gemini and ChatGPT. If the charts you need to analyze are harder to read than this (for example something with many lines in different colors that overlap) ai is going to have a harder time. I don't know what kind of data extraction you're looking to do specifically so the prompt I added with each image was just "Please extract the data from this line chart."

ChatGPT (OpenAI’s GPT-4-turbo model): The initial response was just telling me the labels of the X and Y axis with upper and lower bounds. It then asked I wanted it to digitize the file and extract approximate data points from it and I said yes. It then ran the chart through an edge detection algo and started to run another process on the image but I only have the free version so it said it couldn't complete the task. I then asked it to just do it without the advanced process and it gave me data points but they were completely inaccurate.

Gemini 2.5 Flash Preview: It gave me a table but it made some strange errors, like giving random amounts of days between dates and some inaccurate data. I would not trust it without checking it closely or re-prompting it with more specific directions after making mistakes.

For fun, I also pulled up the chart with the screen streaming on Gemini 2.0 Flash 001. It correctly read the x and y axis information. I then asked it to give me a weekly approximate value based on the chart and it started out accurate but started giving random numbers after that.

With most AI things, it takes quite a bit of trial and error to find the right model that does what you want it to do.

Since you're asking for an explanation like you're truly stupid I will also explain some things I didn't mention earlier below.

ChatGPT is accessed at ChatGPT.com . I just used the free base model that is default on the site now (OpenAI’s GPT-4-turbo model.) Gemini flash models are accessed at aistudio.google.com and they have a bunch of different models (change under "Run Settings" at the right.) If you want to stream from your webcam or stream your desktop (so the ai is actually looking at your webcam image or your desktop itself) use "Stream" from the menu on the left. Different models have different abilities (Gemini 2.0 Flash is the only one with image generation for example) so just play with them to find one that works best for the task you're doing. There are other chat bots available (I used to use llama at huggingface.co/chat/ but the site isn't loading for me today so idk.)

I know I have ai in my username but I am not a developer or advanced technician at all, I use it mainly for creative/art purposes so I'm interested to hear other people's responses as well. If anyone knows better ais to use for tasks like this I'd like to know.

Let me know if this all makes sense or you want me to explain anything better. Your best bet is just going straight to gemini or chatgpt and feeding it images of your charts and see what works for yourself with trial and error.

Dang, my line chart is at least as dense as that and with multiple lines. It sounds like the AI/image-processing just isn't up to the task yet, which I find very surprising given that all we are asking it to do is to recognize where colored pixels sit on a grid of pixels! Thanks so much for your response.

You’re welcome. It’s surprising to me too that the current models I tried couldn’t complete the task. I use ai almost every day for various tasks and it still surprises me nearly every day but half the time I’m surprised at how good it is at tasks that were impossible a few years or months ago and half the time I’m surprised at how bad it still is at simple tasks. That’s why I’m very skeptical of AGI/ASI happening in the next decade personally

I passed the N3! While I probably won't take a higher level test, I will be continuing to build up towards full fluency. Anyone else get their results back? Any suggestions for new anime to watch, podcasts/books to listen to, web-serials to read, or japanese shows to watch?

Also, I finished the first part of an overhaul project porting my personal DMing tool from Netbeans to Excel. The main driver was how clunky the java was but I also lost the central file and only had the executable. Rewriting it in netbeans wasn't going to happen and if I was doing it from the ground up, I wanted something easy to edit and use on other devices. Lots of interesting math and structure, but the basic foundation is using Excel to pull random magical items, groups of magical items, and treasures based on a central, fixed seed. Since I wrote the original, I've changed how I DM in some major ways so I've been meaning to update it for years. Next steps are to add back in the random encounters, rooms, and other things back in. https://anarchydice.wordpress.com/2025/03/16/random-item-generator/

Update on the liqueurs: https://www.themotte.org/post/1259/tinker-tuesday-for-november-19-2024/271334?context=8#context Those walnut and pecan liqueurs? They aged really nicely and mellowed into a delicious, well-rounded, nuttiness. Definitely moved up the list to be right around the brown-sugar oatmeal.

Do you read manga? It’s pretty ecchi but I reallly liked 地球の放課後: the language is straightforward, the writing and plot really surprisingly good, and the art of post-apocalyptic Tokyo beautiful.

Also 海の御先, a harem manga but with again really good characters including MC.

For less moe stuff, I really like ぐらんぶる, a comedy about the Izu University Scuba Diving Club and its collection of (mostly male) morons.

I'll give them a look to see if I'm interested, thanks!

If you find 地球の放課後 nicely drawn but too ecchi you could also try YKK. If you haven’t figured it out I have a weakness for pen and ink ;)

Ok, let's go over the checklist from last week:

  • Make the "subscribe" button work
  • Add a script to download content from the subscribed to profiles - This one will need cleanup / optimization, and I just barely cobbled it together for now.
  • Start on reorganizing the database architecture to optimize performance.

So could be better, but I'll use that Psycho-Pass review as an excuse for why things went a little slower.

How have you been doing @Southkraut?

In between work and kid, Easter. Time flies; feels like last Tinker Tuesday was mere minutes ago. Feels like I'll drop dead, grey and all wrinkles, before I get to write another line of code in private.

Thanks for keeping it up.

Is asking questions about a potential project allowed here?

I want to setup a basic home networking/security lab. I have a bunch of old computers, and I have a few years of experience in IT and cybersecurity.

My goal is to have some stuff I can pentest and just generally mess around with to build my skills.

I think I’m just overwhelmed with the options and the different guides out there.

I have one of sorts.

I just frustrated my children with an Wi-fi outage while I finally vlan'd the wireless and configured the Starlink as a 2nd untrusted interface on the firewall / router.

A good starting-off reference point would be FUTO's guide to a self-managed home cloud setup here (though their wiki is down at the moment, so here's an archive). It's pretty comprehensive so you can kinda pick and choose the parts you actually want to implement.

Not that this really helps with paring down your options and such, but I figure it's well-structured enough that you'll have a relatively easy time parsing it.

FUTO's guide to a self-managed home cloud setup here (though their wiki is down at the moment

"If your self-hosting tutorial brought you to this..."

It will probably come in helpful to equip one of those old computers with dual NICs so that you can experiment with different ways of managing network traffic.

One thing you'll need is a decent switch. You can usually pick up retired enterprise switches on eBay for not too much. But you'll want something that can at least handle a variety of VLAN configurations, so you can segment your testing networks and prevent things from talking to each other that should not be able to.

You might also want to consider beefing up one of the computers (or getting a retired server) with RAM and installing Proxmox, a free hypervisor for KVM/LXC. You can then set up a VM for docker containers, and VMs for various test systems or toolkits.

At the very basic end of things, it seems the easiest way to get started would be Docker or VMs. It's maybe a bit dated at this point, but when I took a CS course on security quite a while back, Google's Gruyere ("Swiss cheese, get it?") was a good toy target application, and there are a number of easily-searchable links for getting that running locally.

I don't directly work in pentesting, so I can't really point you at specific resources, but I think like most folks in tech I've at least had to see the other side of things ("security policy requires these changes"). The concern I'd have for you is that cybersecurity is a rabbit hole both wide and deep: I doubt there are many folks that truly understand all the details of cryptography and implementations (Debian SSH key generation, Heartbleed, Shellshock) and hardware implementation details (Spectre, Meltdown, Rowhammer), or any of a number of other relevant details (rubber hoses). If you just want to try out some fun SQL/JS injection attacks and browser development tools, Gruyere is probably a good starting point, but not being directly in the pentest side of the industry I can't speak to how useful those skills are these days given automated scanning tools for code. I can tell you that I'm pretty careful to sanitize my inputs.

Is asking questions about a potential project allowed here?

Absolutely, plenty of people do it! Anything that will get you tinkering.

Sadly your question doesn't is outside of my expertise.

When last I left you, I had milled most of the parts for my set of chairs, glued up the panels for the seats, and made a template for the back legs.

This week was all about trying to get ready for the weekend when a buddy came over to help me out. This was a fairly repetitive process of doing 96 tenons on 48 work pieces on my router table. I keep thinking it would be faster to do this on my tablesaw, with the proper jigs to support a workpiece on it's end. I should probably try that one day. Doing a handful of tenons on the router isn't bad. Doing 96 of them feels like an enormous waste of time when you know the tablesaw would make quicker work of it. Ah well, maybe next time.

I used a palm router for the mortises in the back rest pieces. It went ok, and since these just hold the slats in for the back rest and are less structural than the mortises in the legs will be, it was sort of an experiment to see if the palm router is sufficient at scale. Pretty sure my answer is no. The mortises came out okish. I had to do them in 4 passes, reaching a total depth of 3/4", and weird things happened where it would grab a side just slightly wrong and rip out huge chunks on a deeper pass, or reach the end of a pass, contact the end wrong and somehow explode a chunk out of it. The nice thing about mortises, I think, is they can not be perfect. Glue does most of the work, and so long as there is enough good contact, they'll be fine. Any imperfections get covered up when the tenoned piece overlaps it. All the same, I want to do better on the legs.

So I got a plunge router from Harbor Freight. Short of a dedicated mortising machine, this is what I've seen used most often for mortises. I got started on a jig for it, which was super easy. Slapped on a fence, used a 1/2" bit to route out a 1-1/2" and 2-1/2" slot, and then a 1/2" bushing with a 1/4" bit will make the mortises easy peasy. I do need to pay more attention to some sort of clamping contraption on the back side though, because when I used it on a test piece, all clamped down to the table, the work piece walked off a huge amount. So some improvements will need to be made before I use it on the legs. Either adding a fence to the bottom, gluing some sandpaper to it so it had a bit more grip against a workpiece, or somehow adding rails to the bottom that clamps can be used with to clamp the piece directly to the jig.

Speaking of the legs, I picked up the 6/4 walnut I needed for those, and when my buddy came over we got to work. I didn't stop at all to take photos cause it just didn't occur to me, but here is the finished product. At least the first one. The template worked pretty well. Has a lot more milling marks than I anticipated, but those will sand out. Took about 90 minutes from rough lumber to semi-final product, which may not be bad? Usually I'd do them all in a batch, but I wanted to take my friend through the whole process on one leg so we weren't just on the jointer all day.

Only thing that went wrong was at one point the weight of the leg caused it to lift off the router table, and the router took a small chunk out of the template. I need to repair that before using it again, otherwise it will just transfer that mistake to all the other legs. I'm thinking sawdust and superglue, or something more temporary, but making a fresh template off it that will hold up better. I may do both just so I have a backup template if it happens again.

Had my first go at using a (borrowed) router today. Need to rig up a ghetto method of deflecting the dust as my lower half looked like a pine-y snowman and there was practically a radiation style shadow behind me after profiling one small piece. Good results otherwise after one quick test piece to better dial in the motor speed and pace of cut to avoid the unanticipated scorching. Far more productive than my one attempt at making a round over using a file but like most power tools despite it's undeniable productivity and accuracy it's not a "nice" tool to use.

Found some crafty YouTube ideas for converting it into a router table too but that's for down the line when I own my own, for now I'm just interested in getting this project structurally finished before the end of the month.

RE: The burning, make sure the bit is clean. I use a spray like this, and so long as you haven't worn the bit dull, it works wonders. I generally clean all my blades and bits after every major project. If you are scorching the wood, that means there is too much heat, which means you are also dulling the blades of the bit. There are some, well not exceptions exactly, but times when this is less the case than others. Maple scorches notoriously easy as I found out the hard way. Just make sure you keep feeding at a steady rate and don't linger for too long in any one area. If you need to reposition, pull the bit away from the workpiece a little and then get back to it.

The other thing is to invest in good bits. In the US Whiteside is a reliable brand that has proven sharper and more durable than anything I'd find in the store, or whatever cheap chinesium shows up on Amazon. I actually grabbed a fresh Whiteside template bit before I began my current project because the one that came in a (I suspect counterfeit) Bosch set was a fucking joke.

I'm also looking forward to building a router table. I have one that came with the first router I bought, and it's ok. I had to buy a new fence for it though, since the one it came with had a problem where the left side of the fence wasn't flush with the right side of the fence. Wrestled with that always taking a slight divot out of the last inch or two of a workpiece for 2 years before I splurged on a $100 "high quality" fence. Was definitely worth it though. Lately I've been desiring a higher quality router lift to build into whatever table I make. Debating building it into the miter saw station I have in mind. But it might have to wait since that's a pricey bit of kit, and this project is supposed to be on the cheaper side.

I think the cutter was new as unlike everything else in the tool case it was spotlessly clean and wrapped in what looked like a factory applied shrink fit rubber cover that unavoidably tore a little when I carefully took it off, so I assume it was its first outing. Good tip though, I probably wouldn't have thought of that. Any suggestions for a non-dedicated cleaner? I've got isopropyl, mineral spirits and a degreaser that I use on my bike chain.

I'm kind of tempted to make TRON style lights for my bicycle helmet, bicycle and vest for myself and my family. A central controller would broadcast what LED should display what color. This way when we cruise around town we can all have synchronized lights and look absolutely sick.

I'm not really worried about doing the Arduino/ESP stuff or the wiring, but I would like to avoid buying a bunch of random stuff blindly and struggling with various LED light technologies. Anyone familiar with what I actually want? Regular very common LED light strips are too dim to use during the daytime. Thoughts?

There's a lot of good information on the WLED discussion boards (and, unfortunately, Discord). If you want to skip the programming side, WLED as a program is also pretty strong, if not necessarily well-documented.

Most of the NeoPixel/DotStar (or simple one-channel RGB 'mood lighting') strips intended for direct Arduino use will cap out around 60 mA per LED, simply because that's the cheapest and easiest configuration for the chips. Adafruit has some ultrabrights, but they're a nightmare of a form factor and probably too bright for what you're doing, even in the day. If you're willing to print out your own circuit boards, getting big LEDs onto a WS2814 chip is an option, but soldering the big heatsinks those LEDs come with onto a flexpcb is not an easy task, even with specialized tools.

Unless you're willing to deal with individual heatsinks on the ribbon, the easy tradeoff is just going with more LEDs in a smaller form factor. You can get individually-addressable ones up to 120LED/m this will have almost four times the brightness of a 30LED/m cheapo strip, along with better resolution. The WS2814s or APA102s are usually going to be the brightest in a given form factor. Going to higher voltages won't get you that much extra brightness, but it will have an impact and drastically simplify the wiring for even moderately-sized runs due to voltage drop issues, so 24v is probably the easiest to do with a 'normal' setup. This does significantly complicate the driver board, though.

Circuit-on-board options would be even brighter, but they're generally not going to have as high a resolution, and a lot of COB boards can only address 'zones' of multiple LEDs (sometimes over an inch per zone, which is how these COB strips advertise such high LED/m values). That said, do look closely at any purchase option; even 'standard' neopixel strips on are often zone-based (cfe here at 120LED/m and 20 zones/m, or here at 896 LED/m and only 16 zone/m). Low zone density will usually result in lower brightness when compared to a (admittedly often theoretical) strip of the same LED density.

Depending on the complexity of your intended LED patterns, this may or may not be an acceptable tradeoff.

WS2815s are always going to avoid zones and can be found in much higher density, but they're individually much less bright (at the benefit of being much more power-efficient) and only go up to 12v, along with having higher passive power draw. If you need a lot of detail and don't want to have to fuck with 24v power, the 300+LED/m strips might be worth looking at. I'd still recommend grabbing a sample unit and making sure it's bright enough.

Okay so if I was willing to forego individual addressing of LEDs, sounds like COB lights would be a good way to go?

I was imagining creating a 3D model of where every pixel was and being able to do a really sophisticated animation based on that (e.g. a pulse starting at some center like my vest and radiating out to the edges of my helmet and bottom of my bike), but there's only going to be a couple of light strips per item (helmet, bicycle, vest) so I don't imagine anything too intricate will show up and I should just consider controlling the color of all strips on an item as a single unit.

Depends on whether brightness or fine detail are more important. I'd probably go with the fine detail option, with what you're saying.

Regardless of what you buy, most are going to be somewhere around 2-5 lumen/LED. You don't have to beat the sun to be visible, but as a comparison most floodlights run around 1200 lumens, and automotive turn signals around 500-900 lumens.

None of them will be visibly bright under full sunlight (compare automotive high-beams at 3200 lumens), and I don't think it'd be practical for any clothing you'd want to wear. I'd expect a good quality thin diffuser will at least change color or perceived shade under a 144LED/m WS2815 setup (napkin math says ~500 lumen/meter post-diffuser) under normal outdoor conditions, but it will be subtle regardless of what LED strip you go with. And if you're wanting that sort of sophisticated animation, being able to control individual squares rather than long strips seems likely to be a bigger benefit compared to just adding additional parallel strips.

(imo, even 'normal' LED strips you won't want full brightness at night.)

There are premanufactured flexible 'sheets' of WS2815s that may be easier to work with for something like a vest, if they have enough density and brightness for your task.

Thanks, good info. I'm going to try ordering some things and see how they go.

There are premanufactured flexible 'sheets' of WS2815s that may be easier to work with for something like a vest, if they have enough density and brightness for your task.

Oh jeez I was just planning to do the contours of our cargo bike with strips but after all, why not? Why shouldn't I paper every square inch of the box with these sheets?

Do some napkin math for power consumption, weight (esp of wiring), and waste heat (est ~60% of LED power will eventually go into whatever surface they're mounted to as heat). It's easy to think of these things as rounding errors, individually, but they do add up, and it's easy for your eyes to get bigger than your stomach when it comes to them. LiPos let you get away with a lot of goofy things from a power consumption perspective, but they're not free.

How do the drivers for these work? Is it the simple series string + a resistor you see in a lot of lighting strips, or does it use a current limited supply?

I've been wanting to set up a DC led lighting system because I'm so sick and tired of the crappy built-in bulb power supplies, but actually driving these things off a varying battery voltage is kinda difficult

NeoPixels (WS28xx), DotStar (APA102/SK6812/SK9822), and most circuit-on-board designs do their own current control for each (zone of) LED -- the only real challenges for implementation are making sure you don't have too much voltage drop (because current control can't adjust for input voltage below the forward voltage drop), and if so just running additional power connections to the middle of the strip. They're not great for room lighting because of the color quality, though, even the RGBW variants, and diffusers only help so much. They're designed for constant voltage drivers; using a constant current power supply can cause problems ranging from comm issues to drastically reduced lifespan.

Cheap RGB lighting strips will almost universally do the series+resistor thing, as will even some decent single-color room lighting. Constant-current drivers and LED strips built for them exist, but you're usually stuck with very specific lengths of LED strip as a result; unless you really need the extra brightness uniformity, I dunno that I can really recommend any.

If you're trying to work with battery voltage to a non-current-controlled output, I really recommend a buck-boost-buck voltage stabilizer. You can get 12v ones for small or mid-sized applications that will handle the full voltage range you want to run a lead-acid battery down to, and are good on output within about 5%. Only downside is that they don't like starting in <-10F cold temperatures.

I got a cheap 10A (<5A realistically) one of those, 24->12. I don't have any way to test ripple, but at least it'll be easy to see if it makes LEDs flicker. Some of the chineseium power electronics is amazing, some of it's absolutely terrifying.

It's weird that for all the problems AC LED lighting has, DC is somehow no better lol. Seems like that's true of a lot of stuff that you'd think direct DC would be good for.

98k words on my NaNoWriMo project. First draft now projected to run to 120k. I'm hoping it will take me fewer than 22 days to get there.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats