HN user
https://jmw.name
The Youtube introduction video (https://www.youtube.com/watch?v=igVscvWAR1s) is a great explanation of the tech and the end where he retrofits it into a Panasonic CF RZ is wild.
TrueNAS is just web-based configuration management. As long as you only use the web UI, your system state can be distilled down to the config file it generates.
If you do a vanilla FreeBSD+samba+NFS+ZFS setup, you'll need to edit several files around the file system, which are easy to forget months down the line in case of adjustment or disaster recovery.
Firefox does do the right thing and seems the most usable browser for private CAs. Chrome and derivatives mostly too, except the problem mentioned about the public suffix list. Mobile clients seem the most broken. I can't get iOS to work well with my private CA packaged into a .mobileconfig, but it could be my error as well.
Another obnoxious behavior is clients enforcing lifetime requirements for domains they have no business imposing their opinion about: .internal and .home.arpa. These are specifically carved out for private use. If I want to roll my own CA with a 2.5.29.30 name constraint extension for one of these domains and hand out a 10 year wildcard certificate, I should be able to without interference from my web browser.
Additionally, Google and the PSL have inadvertently broken .home.arpa on Chrome by misclassifying it as a public suffix, while leaving .internal alone. A wildcard cert for *.home.arpa will not work on Chrome, but *.internal will, despite these two domains being essentially equivalent in purpose.
HyperCard was my introduction to programming and delivered on the vision of personal computing as "bicycle for the mind." RIP
Working for a great company in its heyday is a gift - one that I wish for everyone. Stories like this are a comfort when the industry is near its nadir, and reminder that the industry moves in cycles, and all glory fades. I got my turn at Facebook in 2010. A bunch of times I'd see a name I'd recognize pop up in internal discussions: an esteemed classmate or colleague had joined, and you knew with all this talent concentrating in one place, good things were to come.
The lack of the ohm symbol Ω is also quite a bummer given the technical domain of the font.
I use KaTex (https://katex.org/) as part of static site generation, and I get LaTeX quality output (because it duplicates LaTeX's algorithm) using only CSS.
Using the eqn and troff suite in 2025 ... I'll just say there's a better way. LaTeX is arcane enough, but at least it is a universal standard in mathematical publication that it pays off to learn a little.
I made a test PCB with capacitor footprints repeated at various intervals, with measurement ports for controlled experiments. You can really see the performance difference between two and four layer PCBs, for example: https://jmw.name/projects/exploring-pdns/
Who works Sunday morning then?
The maintenance window will morph into a do-big-risky-changes window, which means everybody in engineering will have to be on-call. Many years ago, when I newly joined a FAANG, I asked, "shouldn't I run this migration after hours when load is low?" and the response was firm, "No, you'll run it when people are around to fix things". It may not always be the answer, but in general, I want to do maintenance when people are present and willing to respond, not nights and weekends when they're somewhere else and can't be found.
Tuning feedback loops is a topic I've found to be fiendishly good fun, perhaps because of how obtuse the final result is. There is no at-a-glance way to see the connection between your high-level goals (closed-loop bandwidth and phase margin) and the implementation (the R and C values). Consequently, I make it a point to document the hell out of these circuits with parameterized simulations so that my future self has some hope of understanding and adjusting them later on.
A lot of applications are usually tolerant of suboptimal compensator design. I've participated in a few designs where these circuits were plucked from the datasheet reference design and never touched. There's a tradeoff between having a little bit of ringing vs. having an engineer model, tune, and test, and also adding to the BOM complexity with a bunch of different passive part values.
they aint ever going to pop a North Korean threat actor bc they simply cant travel at will.
True, but the USG has a long memory and holds grudges. Even if they never travel, they have to be confident every future government of the country will have their back. What's the odds the North Korean or Russian regime substantially changes in their lifetimes? Probably higher than the chance a future US administration will stop caring about an outstanding warrant.
In the mid nineties the Internet was largely a research network between university computers and paid for by tax payers.
Your timeline is off by at least half a decade, and things were changing very rapidly in that time. By the mid-'90s, NSFNET was formally dead, after years of accepting commercial traffic. Local ISPs for home users started popping up and AOL opened its access to USENET in 1993.
The push for residential broadband also started almost immediately; @Home was offering residential cable internet in 1996, it was the future at the time, just not very evenly distributed. This was well before PCs had the processing power to do standard-definition video.
The space of possible subjects is huge, so on average your discernment is terrible, relative to what it could be. This is a serious problem if you create a machine that does everyone's job for them.
It's a problem if you're building systems as a dilettante. There are enough individuals on the planet that people can specialize so we need not be at the mercy of systems built by people of average discernment.
If their art dies out, maybe nobody will know how bad all the pianos are. And then we'll all have slightly worse pianos than we would otherwise have. And I mean if that's the way things are going to go, then let's just steer the Earth into the Sun, because what's the point of any of this.
One counterexample to this viewpoint is Damascus steel. The exact art of making it has been lost (although now effectively reproduced), but modern steels surpass the qualities that made Damascus steel prized. It turns out those ancient masters didn't know bad their steel was. Maybe modern piano tuners don't know how bad their technique is, once the process of "finessing how the overtones interact with each other" has been thoroughly characterized by present or future technology.
I had this discussion (specifically about the Hippie trail) with a friend before I made a trip to Hong Kong in 2019. The conclusion was it is important to travel while you can. Things can change and not always for the better. I returned from Hong Kong on June 3, and less than a week later the protests started and turned violent. Then the next year, global travel all but shut down. I also had the opportunity to visit Kyiv in 2019 and regret not taking it.
Not a true top-line overall number, but the trend is obvious notwithstanding.
Frequency domain analysis is useful when you have a physical quantity (which normally comes down to energy or power, in joules or watts) that you want to understand in terms of frequency. Mere periodicity of events doesn't imply that is the case. Statistical techniques, on the other hand, are widely applicable to time series data.
There's different levels of "understanding how things work" and the author makes it clear what kind of understanding he's going for. If you look at the source code to program, you should be able to point to any line of code and answer "What does this particular line of code do? Why is it important and how does it relate to the rest of the design?" Same applies to a part on a electronic schematic or a mechanical drawing. There is likely no similar meaningful answer to those questions if you look at a particular weight in a model.
(2022)
The first time I saw a PCB business card, I thought it was pretty cool. Then I saw a PCBA business card (with components!) and was amazed. But now I just see them as unnecessary e-waste for vanity. Business cards get read, scanned, and tossed. At least the impact of a piece of paper and ink is small compared to fiberglass resin, copper foil, ENIG, and solder mask. This example isn't even that great as a business card: the typography and contrast make the contact information poorly legible compared to the component silkscreen. Amazing PCB art (https://grandideastudio.com/portfolio/projects/the-worlds-th... is the best I've seen) makes creative use of the different contrast, translucency, and textures between exposed and masked copper, masked and unmasked bare FR4, and silkscreen layers. It's a very constrained graphic design problem that takes a good eye.
A lot of it is not mass-produced (https://www.latimes.com/local/education/la-me-caltech-glassb...). Even if you used brand-new pieces, for analytical work, you'd still need to clean them according to these procedures to remove residues from manufacturing, storage, and shipping.
Using Alconox (mentioned on the page, and easily found on Amazon) is one my tricks when I absolutely need sparkling clean dishes or containers. Not the best choice for the environment, and too sudsy to use in a dishwasher, but amazingly effective.
The amount of written material (work samples, writing samples, free-form essay responses) they request is staggering. There is no guarantee they will read your responses with the same care you put into writing them - almost certainly they won't, since they have actual work to do and an honest reading of over a dozen responses, on technical subjects, is easily a huge chunk of a day. At least with a real-time interview, if you're wasting my time, I'm wasting an equal amount of yours. There is an incentive for the interviewer to get the signal they need to make a decision and respect the candidate's time, since it's their time too. This hiring approach won't scale, for them, or the people they need to hire, if they expand, so at some point they will be forced to change it. Until then, I think the better approach if you're interested in a place like this is to be so awesome they have to call you first.
I ran into this when trying to make a vector version of the Smith chart [1] - which has a lot of circles. It turns out the inaccuracy from a 4-segment Bezier "circle" is significant when you need precisely intersecting arcs at the scale of a full letter/A4 page [2][3]. Ultimately, you can just increase the number of Bezier segments and make the error arbitrarily small; doubling to 8 segments was enough to eliminate the error on the chart even at extreme magnification.
[1] https://www.penpapernomographic.com/pdf/smith-chart.pdf
[2] https://github.com/pgf-tikz/pgf/issues/817
[3] https://camo.githubusercontent.com/0cc44db2e367356629d897c10...
Motors, wall warts with line-frequency transformers, or other things that are an inductive load. Magnetic materials designed for 60 Hz may saturate at 50 Hz, though usually things designed for 50 Hz will at least sort of work at 60 Hz.
1 TB is not a lot, so the answer is probably snapshots to an external hard drive. Together with cloud backups, you have 3 copies of the data on different media and in different locations, which is very robust.
10x that amount and maybe you can start thinking about tape.
Last statement makes no sense since AD and Qorvo are in the business of selling silicon. These tools are marketing for their product lines.
It's like saying Google and Facebook are at a business (dis)advantage over the merits of Tensorflow and PyTorch, when their actual business is something completely different.
The touted benefits of QSpsice include better numerical fundamentals, so who knows what happens when you take a tuned model and plug it into a different engine? The way it is described, the improvements to the SPICE engine could be valuable to anyone interested in numerical code and scientific computing, all the more reason why keeping it closed is a shame.
It's a shame that circuit simulators, especially those derived from SPICE, have tended toward closed-source. It's nice that there are several free-as-in-beer options available, but a good lesson in the consequences of permissive open-source licenses. SPICE predates copyleft licensing, so software that was once open has disappeared into proprietary versions.
LTspice (sponsored by Analog Devices) is closed-source. The recently-released QSpice (sponsored by Qorvo) also closed-source. QucsStudio was taken closed-source by Qucs' original author, and he's not commented on his intentions in this decision.
Big companies typically open-source software when it's not part of their strategic business value, and offers community engagement benefits and good press. LTspice is kind of an advertisement for AD's chips. AD and Qorvo aren't remotely in the software and EDA business, so their decision to keep them closed is even more puzzling, but I guess the instinct of hardware companies is guard everything.
Curtailing legacy admissions would push some privileged kids to second tier schools and cause more of this happen, instead of concentrating it at a small handful of top schools.
The event where some kid of humble origin connects with a kid of the Social Register crowd and forms a lifelong bond comes from mostly odd chance, and it tends to recedes after the initial mix of roommates and floormates are assigned. Social circles gel and people start sorting themselves almost right away.
Kodak & Ilford sell regular selenium toners, and https://stores.photoformulary.com/toners/ for more exotic stuff