OP is Undoubtedly referring to "In Tune And On Time"
Live at the Brixton Academy, 2004.
I have it on a DVD. I think it's all on youtube, but split into parts.
One of my favorite live performance recordings out of many, many, many...
3,757 karma · joined October 10, 2022
Live at the Brixton Academy, 2004.
I have it on a DVD. I think it's all on youtube, but split into parts.
One of my favorite live performance recordings out of many, many, many...
Everyone knows that debugging is twice as hard as writing a program in the first place. So if you're as clever as you can be when you write it, how will you ever debug it?
It's prescient.
Happens on lakes in Wisconsin. See [0] for a very interesting example of what happens when one gets real big and floats away.
I recently had some old hardware manuals I wanted to digitize, which were pages in 3-ring binders. I started thinking about scanning them but did not have an auto-loading, xerox style scanner at home. I looked around for free software that could do this 'google books style' as they did here in this article - take a photo of each page and automatically find the corners, dewarp, white balance, etc. and export to PDF.
I actually didn't find a premade solution that looked like what I wanted. So I wrote my own, with Python an a simple tk GUI.
I tried a lot Open CV messing about to do automatic page corner detection and whatnot, but found everything I tried unreliable. Sure, the corners may autodetect about 95% of the time, but if 5% of the pages fail and get cropped wrong or dewarped weird, I noted the time it took to go back and fix those errors eclipsed the time saved by the entire rest of the automatic process.
The fastest method I found was a human in the loop semi-auto program: it loaded in all the images from SD card, and had the user manually click the 4-corners of the page as shown in the image. Upon the 4th point being clicked, the page would auto-de-perspective and apply the last saved default white balance curve, thresholding, and posterizing (color binning) / grayscaling / scaling to fixed width. For subsequent images in the camera roll this meant settings didn't have to usually be changed on a per-image basis. Then just a spacebar to add the image to the saved temp queue and at the end an export to PDF.
This approach had me digitizing about 600 pages in an hour I estimate. Yes, it was a lot of clicking for the manual part.... but it allowed me to proof every page as it flew by and this caught a few photographic errors that I wouldn't have caught if it was fully automated, and not a single page was left mis-shapen or unreadable.
The problem is... now for anyone looking to pickup a slightly older used/surplus server rack - you have the go and buy the (now marked up) RAM separate.
So... did they list the "0GB" racks with removed ram cheaper you ask? Nope. Many of them seem to be basically the same price as they were or even higher than they were, except now you're paying for RAM separate.
I bought a surplus Gigabyte server about 2 years ago off ebay for a home project, and I just went and the same seller still has some for sale, even older now than they were when I grabbed one, but they are 2x the price, and they don't come with any RAM, whereas I got one already populated as they were from the factory. Once you add that back in literally looking at 150-200% markup in the last 2 years on what was just surplus computer equipment that is 5+ years old.
After the crypto miners dumped them (that was, what, 2 years ago?), I saw one of the entire rack systems on eBay- I think they held 12 of these in one rack on a common backplane - for $800. (or maybe less?)
That's $66 per BC-250.
The real takeaway is that power is power and energy is energy and regardless of how it gets to do the 'pushing' of the air, if you want to use solar energy (power, at any given moment, not integrating over time) those are the numbers you are comparing. It's all just unit conversions, at least when you're attempting such napkin math. Anything more - taking into account the actual systems, losses, efficiencies, etc.. just makes it all worse, not better.
A cool exercise is - given the 777 wingspan I esitmate and the power output from solar of said size... what kind of current aircraft use piston-driven engines with similar horsepower? Assuming we had motor inverter electronics and an electric motor that was 100% efficient, you could imagine trying to build a similar weight aircraft of that size that has such a wingspan.
This is why you wind up with the only solar powered aircraft out there being superlight, high altitude craft with super high aspect wing surfaces- maximizing wing area to weight / lift capability such as: https://en.wikipedia.org/wiki/AeroVironment_Helios_Prototype (note the largest of those prototypes and the tiny amount of power output they were making. All that wing area and it would have barely been able to carry a single passenger as payload and cruise a day without battery power)
Take a Boeing 777, it has a wingspan of about 60 meters, and I'll ballpark an average upper wing chord of about 7 meters, for a total upper flat area of about 420 square meters.
High quality modern but standard single sided solar panels can do about 220 watts per square meter is full sunlight (around 22% efficiency tested at a 1000w/m^2 irradiance).
So that is 92,400 watts at full power.
92kW is less than the peak power of a Nissan Leaf. 92kW is 123 horsepower.
The two GE turbofans of a 777 are generating something like 40-50 MW of shaft power during cruise. MW.... megawatts. 50-60,000 horsepower.
Plastering the wing surface of a commercial plane with solar panels would make up less than a quarter of a percent of the total power it uses to produce thrust at cruise, at best case with them fully-lit.
Fully solar sailplanes do exist (NASA's Helios prototypes are an example) but that isn't anything close to a 'normal' aircraft with any appreciable payload/passengers.
So, no need to wonder.
Which means I just built a demo of a tiny 16 bit windows GUI applet, using emulated VB in Win 3.1 in a browser and got it to run on modern Windows in a shorter amount of time than it would have likely taken me to get the same type of demo up and running natively in Windows 10 today.
There are entire sects of people on the internet that share this incredibly frustrating memory from their childhood. [1]
[0] https://www.worthpoint.com/worthopedia/1995-packard-bell-sof...
[1] https://steamcommunity.com/app/359420/discussions/0/13550947...
I then saved this .exe to a disk image, and then saved that disk image back to my modern mac that I'm doing this on via downloading that in the browser.
I then took that .exe and the only dependency - vbrun100.dll - and put it on a floppy (a real one) and then took that floppy over to a vintage computer I have that has Windows 3.1 on it still (an old 25 MHz Intel486 DX industrial computer I have sitting in a closet with no real use for). Popped in the floppy, copied the .exe and .dll.
And ran the program, no problem! A real GUI windows program, made in VB running in Windows 3.1, inside an entire emulated machine in Javascript in a browser window running on an Apple Silicon macbook in 2026....
Exported and saved to disk and then ran without issue on a piece of hardware made in *1992*!
Now I need to do the reverse... make some software with a copy of VB installed on this old machine of mine and run it on my mac inside the browser.
I'm looking at you, 1993's Journeyman Project....
Did you publish the actual trained model anywhere? I see how in the code there is python for how individual samples can be generated, but the model checkpoint pulldown comes from a directory that... I don't see.
I then went through the code of how this runs on the web and- I'm not a web dev guy- so I'm pretty confused at all the bits bolted together to make this into a web app. It seems like there is a WASM bit that is compiled from a typical C++ audio plugin that is doing the stuff like conv reverb and limiting and distortion in the web app - all that is oldschool, non-generative AI, DSP being applied to the samples. Then the samples are just... a few default generated samples, to start- where are they pulled from, physically? And you have a login requirement to spool up the actual generative AI part to generate new samples to run into the DSP (because that needs a GPU on the backend to do, so, a login to help rate limit this)
How big is the actual generative model? Did you ever think about building the generation engine into the WASM bundle, using maybe WebGPU in the WASM to accelerate in a platform agnostic way, so that the entire app would run offline in someone's browser window?
I'm having fun just playing with the kick program without a login, which, again, am I right in saying in that mode there is no gen AI of sampled happening server side, it is just playing starting with some pre-made samples?
sleep 0.1
Slowing it to a 10hz refresh makes it look great.
Further down that jetty are structures with the actual Dolos geometry. It would be interesting to know if, specifically, those geometries were chosen specifically to be placed in those two parts of the wall as they are for specific properties, or if came down to having certain quantities made on hand and their distributions matched the shape of the wall required, or what.
You can see what I mean, this distinction in the exact location where that photo is taken on google maps, here:
https://maps.app.goo.gl/Vn3tGhM81oMqPVpm9
-33.899126, 18.412751 Mouille Point, Cape Town, 8005, South Africa
it's 'Jevonsian insanity'.
Above that reads this bit:
>Its driver door panel consolidates mirror adjustment, mirror fold, door locks, all four window controls, and child locks into a single networked module. That consolidation exemplifies BYD's vertical integration favoring fewer subassemblies, each designed in-house and dropped into place, with firmware determining how any of it behaves.
But this whole article is written from a weird authoritative viewpoint when really I think it should back down a bit and just describe what the damn CT scans show.
>Fourteen pins in two parallel rows carry every signal this panel produces to the rest of the vehicle. Automotive connectors are among the most common failure points in modern cars: corrosion, fretting, and thermal cycling work on these joints over years of use. One connector failure on a module this integrated takes out mirrors, windows, locks, and child safety all at once.
Corrosion on an interior module connector is not as much of a concern these days unless the car is in a flood or the door card sealing is broken due to something like a poor repair job.
Fretting? What would cause fretting on pins of a connector that never gets touched by a human after it leaves the factory. It is a static connection, it doesn't get plugged and unplugged.
Thermal cycling? It is inside a door panel... not near a hot exhaust or inside an engine. It sees normal interior temperature cycles.
An actual Closures Engineer would more likely call out vibration shock during door slam in a closures FMEA as a potential electrical window switch fault hazard resulting from the connector loosening if the chosen connector lacks sufficient mechanical fastening moreso than anything..
Saying that connectors themselves are "among the most common failure points in modern cars".... just sets of flags to me as overly flatulent, generated puff writing. "Oh I need to list three things about connectors (thinking).... Corrosion -Fretting- Thermal Cycling-!! and this makes connectors among the most common failures in modern cars (no sources cited)".
(I'm a former automotive engineer.)
Electrical connectors on automotive harnesses are far, far more reliable now on modern cars than they ever have been in the past, even with the increase in number of such connectors.
Companies like Delphi and Amphenol put immense engineering effort into the way modern connectors are designed, with weather sealing and contact plating that is way better than anything pre 1990's-ish was. Plastics really got way better after the turn of the Millenium compared to what they were before- I remember working on 1980s cars in the early 2000s (so, 20 year old cars) where connectors would ofter crumble in your hands when disconnecting them, or have completely corroded terminals, terrible sockets that yield and wouldn't keep continuity, etc. Compared that to all the cars I've worked on in the last 20 years where connectors have just not really been a worry (among the brands I've work on, at least). The electronics / sensors / modules themselves are much more likely to fail than the connectors they attach with, in my current experience. The only time connectors seem to fail is when cycled roughly/wrongly by people doing service incorrectly. Failure in place? Rare.
Anyone who thinks automotive electronics were more reliable in the past, live a bit with the electrical systems of a car from the 1970's thru the mid 1990's and report back to me on how that goes. Bonus points if it's Italian or British.
>All employees want, in theory, to take as much vacation as possible. But they also all want to take just slightly less vacation than each other, to be perceived as more loyal, more committed, and more dedicated (hence more promotion-worthy). Everyone looks to the others for a baseline, and will take just slightly less than that. The Nash equilibrium of this game is zero.This gives enough friction to the point, when I muscle-memory type the URL in a browser, and get an unreachable error... after a while I learn to just... break the habit.
and so now I'm wondering how cool /fast / compressed a diffusion image generator could be if the images it was trained on / space it worked in was limited to 1 bit (Floyd-Steinberg / Atkinson / your favorite algo here) dithered images.
Training would surely be pretty quick and probably fit onto one modern GPU.
The simulation is so simplified that I see no terms for the control of pedaling. Riding a real bicycle isn't just about steering and leaning a bit. You need to propel the bicycle a certain amount.
The paper buries this in the following:
>Although the two-neuron network controller works well for a range of speeds, one thing the controller does not do is to try to dampen the instabilities that can arise when riding too slowly or in too sharp of a turn. (This would probably require a third neuron that isdedicated to this task.)
It just led me to finding this:
worth a bookmark.
There are mechanical engineers out there who can literally model objects nearly as fast and they can 'think' about the layout of said object. If you look at the complexity of, say, a CAD model from a real, highly complex aluminum casting section of an automotive subframe, or the living-room-sized cross-fuselage spar forging of a fighter aircraft, with hundreds of ribs and fillets and features- and compare that to the simple model you are trying to make in OpenSCAD, you should quickly realize the parallels in difficulty you are trying to express (similar to the person without knowledge of C++ or Python watching someone be able to build applications by typing code from their fingertips as if they already knew what to type...)
You are struggling for a few reasons- 1., it is a knowledge hurdle of an entire field you are trying to surmount- again, go watch someone actually model a real, complex part and watch the speed, they can do so in a tool like Solidworks, CATIA, NX, etc... at a rate that is far different because they have experience that it can honestly take even good people years to accumulate - and 2. they are using professional tools - you mention OpenSCAD, like it is CAD, but it isn't. It is programmatic mesh generation, and it turns out that programmatically typing out how to generate complex things is much more difficult than a combination of a graphical GUI and graph-based generator that all big CAD programs figured out starting in the 1980s. If those tools you use were really the best way to make complex models 'paramaterized', then why do we design our fighter jets, Formula 1 cars, or Space X rockets in Dassault's CATIA or Siemen's NX ?
You want a LLM to take a sketch to your CAD, but what I'm saying is, there are people out there that can skip the sketch and build the CAD as fast as you can likely hand draw the first sketch, and these are skills you can actually learn, but you may just be using the wrong tools and have not had the practice necessary.
Windows XP's level of 'plug and play' for devices/drivers ushered in the modern OS feel from a usability standpoint, but from a 'get-shit-done' GUI and responsiveness standpoint Win 2000 (and up to Windows Server 2003 by extension) was all I ever wanted/needed.
These may be rose tinted glasses though, and I'd be interested to hear counterpoints.
But by using the LLM to generate code like an SVG graphic is made up of, and then using a rasterized image of that SVG as an input to the diffusion model, this takes place of the raw noise input and guides the denoising process of the diffusion model to put the numerical parts in the right spots.
The LLM is putting the SVG in the right order because the code that drives the SVG is just that - code - and the numerical order is easily defined there, even if it has to follow something like a spiral.
Edit: although LLMs now also may be using thinking modes with their feedback during generation to help with complex positioning when drawing something like an SVG, as I just asked claude to generate me one such spiral number SVG and it did so interactively via thinking, and the code generated is incredibly explicit with positions, so, that must help. But the underlaying idea to two-step SVG-to-diffusion model is the real key here.
Its a transformer, with a CNN refiner after. Specifically, a ViT using the Hiera architecture (https://github.com/facebookresearch/hiera)
The Hiera ViT has dual decoder heads, one for the alpha and one for the RGD foreground, and then a small CNN refiner network to solve some artifacting in the output from the Hiera model.
I'd be very interested to see a long form tech talk of Niko explaining his process of learning ML ropes and building this model.
If the PIO pin could drive a fair amount of current at 3.3v into a long enough wire at that frequency you'd start to get into milliwatts, and AM radio is NOT a band that even amateur license operators can broadcast over a a certain power on. FCC part 15 dictates no more than a 3 meter antenna for personal devices at AM frequencies which is what does the power limiting essentially.
The harmonics fall off quick enough on such a setup that it wouldn't really be a problem - but the only way to really KNOW that is to have a real solid understanding of how this 'radio' you've just made is working, meaning how that square carrier wave is really being driven off the PIO pin, and thus you need the requisite EE knowledge and/or ham radio test equipment and experience.
I've seen more and more of these 'ChatGPT coded up a radio transmitter' posts and it kinda rubs me the wrong way. I'd like to see more calculations and disclaimers for people showing some responsibility with radio, and if it drives people to studying and taking an amateur radio license test that would be for the better...