jerich

I’m the same way; I feel like Claude is doing more than just writing code, it’s getting me unstuck.

I’ve been pulling projects out of the closet that have been sitting there for years. It’s because I can sit down and get started so seamlessly. Before, I might waste the first couple hours configuring my environment and tool setup, but with Claude Code I can just jump in and start building, start solving the real problem.

I just built something this week where I had the parts sitting in my closet for a couple years, but just got curious to see how Claude does with embedded C, so it got me started. (Turns out Claude scanned my drive and found an abandoned C project that was outside my usual DEV folder, and just built on that). The code was 5% of the project, but it got done because Claude Code gave me the momentum push.

For my personal projects, the last 3 weeks have been more productive than the last 3 years.

I hacked together a tech demo of a personal project to add a Jetson Orin Nano to augment the Nikon autofocus system. (Camera->HDMI->Jetson processing->USB control->Camera)

https://github.com/jerich/jetson-face-af

It was a personal project to let Claude Code loose and have something to talk about on LinkedIn, hopefully to start a conversation about how to add some more advanced, more personal, functions to the powerful AF systems out there.

“Nikon AF does a great job of recognizing faces, but it doesn’t know which faces I care about.”

But I wanted to augment, not completely take over the camera; keep the Nikon shooting ergonomics intact.

Even the latest and greatest cameras will lag the processing power of something like a Jetson Nano, or even a mobile chipset, and cameras are meant to have lifetimes of years, so I think a smart camera manufacturer (hopefully Nikon) should add an easy external processing loop to let users add some extra smarts and automation.

It sounds so dystopian, but as someone who’s been searching for a while, I would enthusiastically use a site that charged $5 to apply to a job, with a guaranteed human review. I would target the most relevant postings, make my case in a cover letter, and signal to the hiring team that I believe in my qualification. But mostly, I’m paying to keep out the “I’ve sent out 1,200 resumes with no luck” people.

I guess it’s a shortcut to part of the advice in the article—requires realistic focus and targeting, but skips the cold emails.

This gets such a huge thumbs up that I had to scroll up and reread it to make sure this wasn’t my own post from a revived thread from last year!

I’ve been using a pixelblaze with a long string of cheap 2812 LEDs on my Christmas tree for three years now with tons of compliments from neighbors.

I’m an embedded software guy, and every year I mean to dig in and try roll my own, or do something clever with an RP2040 board (also a shoutout for the Pimoroni Plasma), but the demands of life and “get the light show started” mean I keep using the Pixelblaze.

I even upgraded to their newer versions last year, and used some of the smaller ones to make some LED tutus for my girls that synced pattern with the tree (the tutus were synced with each other for a Christmas show, but it was trivial to then add the tree for fun afterwards).

The mapping is huge for the wow factor, and the pixelblaze makes it so much easier to get something fast and good enough.

There’s so many community-shared patterns to choose from, and it’s been easy to make small modifications to look better once mapped to a tree, though most work as-is.

My project I won’t get done this year is to try to make some calibration patterns and use ChatGPT to analyze some photos/videos to make a 3-D map, but I’ll realistically probably end up with the vaguely-triangular 2-D map again; I can get it done in about 30 minutes now.

The following is a couple years ago. I think last year I was up to 1100 LEDs and the mapping was a bit better, but I didn’t take good videos.

https://youtu.be/hu-RQx_NpAY?si=BMYbafbPAn2XAlU9

Apple should hire a couple hackers to create “end-of-life” firmware for their obsolete devices; give them new life as super-specialized devices. Part green program, part customer delight, even some wacky art projects.

Maybe if an iMac doesn’t have a video input—have it boot as an AirPlay-only monitor.

I’ve got 2 old EOL appleTV boxes sitting in a drawer—again, one last firmware update to make them dedicated AirPlay receivers.

Take my 2011 MacBook Air and make it a dedicated Notes machine/word processor—all it does it run notes and sync with iCloud.

Obsolete iPad picture frame is an obvious one.

They can work on the “Reuse” side of the 3R’s of waste reduction (with reduce and recycle, right?)

PS, I’m available, 9 years embedded SW experience ;)

Anyone who’s read the short story “The Billiard Ball” by Asimov would have taken it into account.

I think they’re over in line behind “The La Brea Tar Pits”

So the US sells some bonds, then slips a couple trillion into a brown envelope to buy Baja California from Mexico, then “lease-to-own” it on very favorable terms (0 down, 0% APR, 150 years) to New Taiwan, or Formosa II. Uproot TSMC and an entire culture and move them over, then let China wave the flag over the mound of dirt left behind. Pencil them in to NAFTA and watch the chips flow over to the new iPhone plant outside Mexicali. Tijuana to Ensenada remains Mexico and is a buffer between them and the USA, and the night markets there are the envy of all food-eating people on the planet—-I can’t wait to try the albondigas soup dumplings.

I think AirBNB has the infrastructure to upend the Realtor market. If they just add an “AirBNB listings” buyers can search and scroll through pictures. They’d have to show exact location, but otherwise pretty much the same as what’s there. Seller pays an upfront listing fee, gets referrals for staging and photos if they want to stand out.

Simply add a “30 minute viewing” rental for $25-$50 (AirBNB keeps $10-25), available to registered users (with already-verified ID). Maybe rent the homeowner a bundle of cams for viewings for extra security.

If a buyer is really interested, have an above-market-rate nightly rental—I’d love to spend a night in a house before paying a million+ dollars for it (without the cams, of course). Maybe half refunded with an offer made thru AirBNB Listings; full refund with an accepted offer (again, less AirBNB fees). Maybe partner with Redfin or Zillow to set prices, provide school and property info, and finish the deals. Do it all cheap enough to make it worth sellers forgoing an MLS listing; should be a flat fee, a tiny fraction of 6% in most cases.

A startup couldn’t do it, but AirBNB has the name recognition, and the pieces already there, right? I’m available for consulting, or send me a fruit basket if it all works out, a nice one, without a lot of melon.

I keep intending to reinvent my own controller with an RP2040, but life gets in the way, so for the second year in a row, I’m using a Pixelblaze to control the 950 LEDs on my Christmas tree. It’s a continuous string (well, 19 strands of 50 lights connected), put on the tree in a zigzag pattern and manually mapped to a vaguely triangular 2-D shape. The patterns are mostly as downloaded from the Pixelblaze repository.

I’d highly recommend Pixelblaze for getting a fairly complex setup working quickly.

https://youtu.be/hu-RQx_NpAY

Although I know financial markets are extremely complex and driven by human psychology, I can’t help thinking about thinking about it in terms of classical control systems.

Do you want a smooth input, or an abrupt step input? Psychologically, I’ve got to believe there would be a quicker response by consumers if rates shot up overnight vs “boiling the frog” with a gradual increases.

In simpler dynamic systems, you get a faster response with overshoot and ripple, maybe this would be similar? But would it be better? I wish the article author had been able to find more discussion by economists talking about the potential effects.

Could you pull out of a recession or pop a forming bubble in months instead of years? Maybe it could lead to a smoother Macro-macroeconomics. But maybe the overshoot is too much; maybe the system is just too chaotic to control effectively.

Financial markets dislike abrupt changes, but I think if the central bank was perfectly transparent about their goals and responses (“To maintain a annual growth rate of 1.9%, Fed decisions will be made based on a PID controller with the following parameters…”), some smart financial engineers should be able to account for the rapid changes in their own models. Maybe there’s even some profit potential anticipating the overshoot and ripple that could provide some dampening effects.

I’m sure I’ve completely Dunning-Kruger’d this and millions of lives would be ruined, but since I’m not the Fed Chairman, we’re all safe.

After growing up on a Mac SE and going off to college with a Centris 650, I went through a string of windows PCs until the 2013 MacBook Air brought me back into the fold.

I still use the 2013 MBA almost daily—mostly as an external keyboard and screen for my iPhone notes, but it really slowed down with Mojave, so I’m trying to decide whether to pick up a cheap replacement M1 model now, or wait for sales on the M2. I think I can hold out a little, but won’t quite make the 10 year mark.

BTW, the c. 1993 Mac Centris was able to boot up last year (at 28 years old!), and thanks to an Ethernet adapter I dumpster-dived from work in the early 2000s, I was even able to get online with it without any extra configuration. Netscape Navigator doesn’t do too well with modern websites (and probably made for a baffling entry in some server logs), but I could at least load the Dole/Kemp ‘96 site.

For the first question, I ended up using this trick for a video editing app on Android phones about 10 years ago in order to cut video together on low end, out of date (Android 2.3) phones. They couldn’t handle video compression in a reasonable time and we didn’t want to upload/download to process on a server.

The point of the app was to sync the cuts to music cues, so each clip had a defined length. I ended up doing it all through file manipulation. You can cut into a video file starting at any arbitrary I-frame then trim it to the desired length. I would cut the input videos down to size then concatenate the files, replacing the audio with the new soundtrack at the end.

It worked great, only took a few seconds to create the final edit. Of course you couldn’t overlay text or filter video, but I still think it was a valid solution.

With the requirement of starting each clip on an I-frame, there was some imprecision in where your cut would actually start—an arteur might have a problem with their masterpiece being butchered that way, but it would certainly work well for some special cases like efficient distribution or being able to show a diff that a video was unaltered outside of timing cuts.

I thought it was most interesting to see how clearly crocheting was “generation-skipping” with the 40-year periods between peaks in popularity: 1890s->1930s->1970s->2010s

Maybe the latest peak is pushing out due to having kids later in life (i.e., grandparents now 50-60 years older than grandkids, rather than 40-50)

The best Olympic coverage I’ve seen was the NBC HD broadcast of the 2002 Salt Lake City winter games. HD was still pretty new in the US. They didn’t have many HD cameras and they didn’t have an HD workflow integrated with the SD broadcast, so it was completely separate from the “overcoming adversity USA athlete” coverage.

For example, they set up the HD cameras and trailer at the ski jumping venue. They showed ski jumping start to finish—all the competitors real-time without editing. The commentators were ski jumping experts, since the professional commentators were all off on the SD feeds. Almost no breaks because they only had one HD commercial.

I learned so much about the sport and cheered for competitors from random countries. It was miles beyond the standard NBC coverage and probably even better than watching it in person.

20 years later and it’s the most memorable Olympics-watching experience of my life, going back my earliest memory of the jetpack in ‘84.

What about 3-D printing for backyard hardscape for higher-end consumers, not homes for refugees?

Mass customization for benches, sculptures, playhouses, fountains. The value is in getting something unique, not speed or cost.

Or does that only make sense after my getting a $4k quote to build a rectangular concrete bench in coastal California? (Actually, I just went back and checked—it was $14.5k for an 18 foot concrete block bench with a plaster finish to look like poured concrete, with an actual concrete version even more expensive).

I made an outline of a business doing this a few years ago: sell yard printers, proprietary media, and object library digital subscriptions to contractors, along with training/certification, but I never went down the rabbit hole to figure out a cement mixture and scale up my Ender-3.

Take a look at what Teledesic was planning in the late 90s. Elon’s a great guy, but nothing I see about Starlink is innovative over what they were working on over 20 years ago.

After a month of Whole30 eating a couple years ago, I've completely switched from dairy in my coffee. Previously, I always used sugar and half-and-half, but a month was enough to make a permanent change.

My favorite by far is unsweetened cashew milk, followed by hazelnut and coconut, but I'm using it as a creamer or latte substitute, not looking for foam. I can't stand the nutty grittiness of almond milk in my coffee, but I now generally prefer nut milks to cow milk and cream.

My first job in 1998 was working on the integration and test plans for Teledesic’s Optical Inter-Satellite Link Tracking (or OISLT, pronounced “oyselt”—probably the worst acronym I ever used). The design wasn’t done yet, but we were already planning integration tests to prove them out before anything was launched.

We were planning on tracking satellites in adjacent orbital planes (which were orbiting in the opposite direction) as well as in-plane. So, yeah, moving parts.

Whenever this topic comes up, I wonder what happened to all the work we did at Motorola in 98-99. There were several hundred engineers in Tempe generating thousands of pages of design documents for over a year. We archived all our work in spring 1999 sand were all shuffled off to other projects.

This article and its predecessor made me think about smaller-scale solar, and I actually bought a 100W panel out of the Amazon gold box last week to start playing around myself.

My first plan is to use the panel and battery to run a set of 12V led outdoor lights, but I'm also thinking about trying to tie into the existing 12V system that I never bother turning on.

That's got me thinking about taking it a step further and adding some capacity to power my Christmas lights this year. Now that they're LEDs as well, it seems it should be possible, but probably not cost effective.

I keep thinking there must be a decent market in between tiny panels powering path lights and a full rooftop system. Like a standalone patio cover or pergola that has solar panels for a roof. That should give a decent amount of power, and it could be an easier sell to people instead of permanently attaching something to their roof.

Actually, I was trying to think of a way to run a small air conditioner from solar—like one of those split systems. The AC could run when the sun is out, which is when the upstairs of my house gets hot. It would be a supplement to my central AC, so I wouldn't be worried about it not running at night or cloudy days. So, I don't need a lot of storage or even worry about tying into the grid.

Again, though, it's probably not cost effective piecing it together, but still the solar patio cover seems like something that should available in a big box at Home Depot or Costco these days.

To be fair, the tinker board is only 50% more and does include additional features (some extra speed and GbE). The US release is happening right now, so that could affect availability worldwide as Asus tries to launch with sufficient quantities. I somehow was able to order one, sold by Amazon US no less, and get it last week; there's not even a listing anymore.

Not too bad for a product that was announced late January.

Ahh, memory lane. LEO satellite internet is one of the subjects that bring me out of lurking.

My first job out of school in 1998 was writing test plans for the antenna pointing and tracking for the Teledesic CPE at Motorola. If I remember right, the initial plan was to have dual steerable dishes inside a pair of radomes—like a pair of Mickey Mouse ears. There was a vague future plan for phased-array antennas, but at the time it wasn't realistic to talk about getting them on roofs at a reasonable price point.

The bigger challenge was the optical intersatellite links; routing traffic between LEO satellites using lasers. The lasers had to track satellites in adjacent planes of the constellation that were traveling in the opposite direction.

It's interesting to see the same general idea for the large constellation come up every few years. If we had actually started building it in 1998, it would've been on a second or maybe even third generation by now.

I'm a little surprised that Apple hasn't given it a try with their billions sitting around. 4K FaceTime, or even just downloading movies or apps instantly would be a selling point for their other hardware.

Wow. What a great question! I found that I was sitting here answering it to myself. After starting down a rabbit-hole on how the browser would probably need to do some GPU initialization, I realized how much fun I'd have talking through this with someone in an interview.

"Let's see, I guess the wifi radio is going to have to do an RSSI measurement..."

It actually gave me goosebumps when I realized that I was so caught up in it and how excited I was getting.

It took me a few looks at this picture to pick up on the fact that it is all in focus, whereas the depth of field of the f/1.8 lens is going to give you one of the subjects' eyes in focus.

I was a little disappointed in the amount of noise and the noise reduction in this pic until I started thinking about how it would look with other cameras. Probably a blurry, noisy mess with cellphone or compacts. To get the same effect as this pic with a DSLR, you'd have to massively stop down that 50mm lens and would end up boosting the ISO so high that even a modern full-frame DSLR wouldn't be that much better.

If you take a look through the FAQ, they do say that they are designing it to work within common photography workflows, giving example formats of JPEG, TIFF, and raw DNG (though I'm assuming the DNG is after their proprietary postprocessing, not a set of the true raw sensor outputs). They're asking for input going forward, so take a few minutes to express your concerns about workflow and maybe it will be a better product for it.

I believe they are doing quite a bit of stacking, as they say they are processing up to 130 megapixels from 10 sensors to get the final image. There should definitely be a dynamic range improvement. I hope they can somehow apply their processing to video streams and provide a wide dynamic range (maybe even HDR) video, but I'd be surprised if they can do more than provide a cellphone-like 4K stream from a single sensor.

Focus speed shouldn't be an issue. I'm guessing it's using tiny cellphone-sized camera sensors, so even at the f/2.4 printed around the lens, they'll have massive depth of field. It looks like they even have the DOF printed on each lens: 10cm-infinity for the 35mm lenses, 40cm-inf for the 70mm, and 100cm-inf for the 150mm. Any focusing will be done computationally at post-processing time.

I'm interested in the concept; I just hope they're a little more developer-friendly then Lytro turned out to be. I know there's a lot of proprietary secret sauce going on with the computational post-processing, but if they can open up their system just a bit to let the crowd in to poke around, there's probably a lot of unique creative opportunities.

It was actually a wife of an executive on a cruise, not a yacht. It looks like you can get cell connections on a cruise now, but I think it took quite a while. Remember, this "idea" was born around 1990.

If you haven't looked, it's actually quite reasonable now to rent an Iridium phone for a backcountry trip these days. If I was taking some boy scouts into the wild for long weekend, I'd definitely consider it in case of emergency.

The system was really a great piece of technology; MOT really let the engineers go at it. If it wasn't for the billing and provisioning, you could actually make an Iridium-to-Iridium call anywhere on the face of the earth in some sort of doomsday scenario, since all of the call routing goes between satellites—you don't need a ground station.

When I started at MOT, we were working on the broadband followup, Celestri, which turned into Teledesic. Same deal, anywhere on the globe, but around a 1-2Mbps connection, rather than Iridium's 0.002Mb data rate. It was a magnificent piece of engineering with hundreds of engineers working in a fancy new building with plans in place to double the building in a couple years as hundreds more engineers were coming on for the detailed design phases.

We were about 9 years from operation and had thousands of pages of architecture and design documents. I think the highest-level system design document was around 500 pages when it was shelved, but there were already some architecture documents running into the thousands of pages.

The projected launch costs alone were around $15B; that's not including actually building anything. It was the most expensive commercial project ever undertaken, though after Iridium's disappointing launch, even the most short-sighted executive could see that it was too big of a risk for any company to take.

Looking back, it was an amazing experience to be a part of a project that huge.

I don't know about his other claims, but this one seemed a little suspicious: "Next I gave him a face by mashing up three portraits from Google images using a free trial copy of Adobe’s “Lightroom” image manipulation software."

As a user of Lightroom since version 1 in 2007, and spending many hundreds of hours using it as my main image-editing program on 40,000+ images since then, it would be my last choice of imaging software to "mash[] up three portraits"

I'm not sure if it's even possible to combine multiple images in Lightroom; it's more for making global adjustments to single images. There might be a way to do it using a third-party plugin, but most Lightroom functions work on the whole image (adjusting exposure, white balance, etc). There are some cloning and healing tools, but as far as I know, there's no way to clone a nose out of one portrait and paste it into another.

Even if it's possible, it's not something that a first-time user would be able to do in the author's two hour window on April 14th after downloading the free trial, especially given all his other activities.

Maybe the author is just trying to CYA for using someone's picture he found online, but if that part is clearly exaggerated, maybe he's just trying to become "internet famous" by making up a made-up "internet famous" alter-ego, or something more meta- like that.

From Wikipedia on the Bayer Filter: "Bryce Bayer's patent (U.S. Patent No. 3,971,065) in 1976 called the green photosensors luminance-sensitive elements and the red and blue ones chrominance-sensitive elements. He used twice as many green elements as red or blue to mimic the physiology of the human eye. The luminance perception of the human retina uses M and L cone cells combined, during daylight vision, which are most sensitive to green light."

The Foveon sensor does sample RGB at every pixel by using 3 layers, but it is only used in a few cameras.

The green channel has the least noise for Bayer-sensor digital cameras (which is almost all of them).

50% of the sensor sites are green (25% each for red and blue), so there is more signal given the same amount of noise.

I really like the idea of this type of physical office layout. I noticed the Mozilla office picture at the bottom of this post the other day: ( https://news.ycombinator.com/item?id=5393864 ), and it really stuck with me. I forgot what the article was about, but kept it open in a tab for a couple days to keep looking back at that picture.

It really seems perfect for everyone working on laptops, but I think most people are more effective with an additional screen. Ideally, I would want some sort of docking station or at least a big monitor on an arm at each booth/table.

I guess the other downside is that it promotes very poor ergonomics. It would be a great setup for a change of pace, but I think if I were looking at a couple years of working in that environment, I'd start seeing some health problems. I guess if you're 22, you can handle anything for a while, but ergonomic mistakes accumulate over the years and can really start to cause a lot of pain. I think I'd love it at first, but after a few months, I'd long for an Aeron chair in a cube again.

Does anyone work in one of these type of offices and can you speak to the short-term and long-term comfort?