I can relate to that. But as I’ve aged I’ve appreciated just how much was done with less back. I mean that literally: we couldn’t afford a computer so my dad would haul home his work computer each weekend. We’d call BBS’ on it, print stuff. Play shareware games. It pissed my mom off because it took up half the dinner table. But my brother learned to code this way too.
So later on when my friends got EGA and VGA, CGA meant “poor.”
Fast forward many years and I have kids and were in better financial circumstances than my dad and mom back then. My kids have MacBooks. One of their friends was over one night and was visibly embarrassed that all she had was the school assigned Chromebook. She and my daughter were doing homework and her friend said that she wished her family could afford better laptops.
I didn’t want to embarrass her but then next morning after I made pancakes and I was sitting at the table I was like hey let’s take a look at these Chromebooks (we had an old one sitting around) and the three of us sat there and in about an hour and a half they’d put together a very simple farm game in the spirit of Stardew Valley on PICO-8.
Point being, we live in a world where especially with AI, it’s easy to think we can do everything. And it can feel like that, which means that when you can’t, you feel worse that you can’t because it would appear everyone else is doing amazing things with no limits.
But maybe we need more limits on things. Maybe that 4 color palette forced devs and artists to be creative back then in ways that make things better for everyone.
Anyway, just on to of my mind: - Indiana Jones , especially the one with the Paris underground and the Nazi fortress. Forgot the name.
- Monkey Islands
- Phantasmagoria 2
- Baphomets Curse 1 and 2 , especially 1.
- Kings Quest 7 (god, the drawings. Like a Disney movie)
- Simon the Sorcerer
All bangers!!!!!
So many crazy memories. They all had really good stories, a nice humour, tough puzzles and really good drawing. Idk, I really can succumb to nostalgia here :-)
Respect for the graphics artists in general for that time to be able to make something out of almost nothing, four colors or less and/or black and white low resolution.
Edit: sorry, i went straight to the paper. I agree, the article itself (and the headline) are bad (as they tend to be on this topic)
> The researchers found the energy and accuracy disparity by measuring the amount of energy used by three open-source AI models, Alibaba’s Qwen 3, Google’s Gemma 3, and Meta’s Llama 3.1.
I've really enjoyed using it on paper: https://hughpyle.com/ASR33/
Terminals work so well because there's a huge frictional cost to adding to the protocol. That's what keeps the protocol simple, which is what keeps multiple implementations feasible. Unlike the web, where unless you have GoogleBux you can't afford to implement it properly.
I don't generally like TUI's and think one gets more flexibility with raw text interfaces and being able to glue things together is more important then them looking nice. I think less fancy out put is better to be honest.
I've written about it before, but in brief the mountain of standards is worth ripping out wholesale and doing write as a semantic protocol. It's needed to solve several things that work really poorly in terminals because their unit of compatibility is cells rather than a higher primitive. It's hard to bake something on top that works well.
The need is most strong in Accessibility, Text wrapping, Scrollback handling (i.e. making the scrollback editable and not dead text), and various other bits that make writing highly functional TUIs. Coding agents tend to find all those edge cases.
This often gets push back that it's easy, to which I can point nay sayers at the various many months long issues on things in Codex, Gemini, Claude etc. on flickering, scrolling, key handline, IME, ... doing anything moderately complex involves modeling a terminal in a terminal app effectively.
Ghostty and Rex sort of do the opposite bit. If you think of it like a stack, they write a new layer underneath that transports the cells on a better multiplexer layer 0. I think a new semantic layer 2 is an approach that would be neat to work with.
I can only speak for myself but I use the CLI and TUIs specifically because of the limitations those modalities impose on software that uses them as the primary method for interfacing with the end user. If you want a full featured UI and all that comes with it why not just use QT or GTK or modern HTML or etc?
I think the only "improvements" I'd like to see are thing such as metadata regarding what the different streams are for, an official way to communicate using structured data as opposed to raw bytes, that sort of thing. Spiritually in the same ballpark as the difference between HTML before and after CSS.
Unfortunately mintty is the only terminal emulator I'm aware of which supports overstrike [0], as this would open up a world of new possibilities for TUI apps if more widely supported.
[0] https://github.com/mintty/mintty/wiki/CtrlSeqs#overstrike
Personally, I don't mind AI content at all when it's disclosed and I choose to participate.
But I regard it as betrayal when I expect human effort on the other side - which I do when I listen to music, watch a video on YouTube, or read a book.
AI is particularly timed in a world where alienation was already a hot issue. I think human curation and human trust networks will become more relevant for people seeking genuinely human things. I know I already appreciate them more.
The AK47 is an invention. The inventor "would prefer to have invented a machine that people could use and that would help farmers with their work - for example a lawnmower", but people still consider it an invention. Inventions do not have a polarity in the axis of quality of life.
There is no consensus on if GenAI is overall harmful or not. In general people are optimistic but concerned.
The last thing we need is another war on drugs.
I would rate A/B testing on human behaviour, or advertising to be more harmful to the overall human experience than GenAI.
Off the shore of Italy, in Montecristo, the Benedictines have regained full control of the island. No Internet or AIs are allowed there, but their hospitality business rents mechanical typewriters by the hour. More than once, a famous writer has jumped from the island's cliffs, unable to bear the shame of being accused of using AI to produce their magnum opus.
For the unwashed masses, AI-produced literature is free and abundant. The discerning, however, have gotten a taste for manually bound tomes exclusively inked with the output of human brains.
This is just one of several areas where GenAI has been harmful and should start getting regulated like digital drugs.
Do they really? They take a cut; they have incentives to let people get scammed.
That's a book a day (give or take), per account. No human author is publishing a book per day.[0]
It's interesting to me how little the incumbent platforms for digital content seemingly have been able to do in combating the tidal, exponential shift in submissions.
I just spent a month developing a Shopify App, on the side, for fun. When I launched it a couple of weeks ago, I discovered an app doing the same thing as mine with an identical name had also been approved and released a week prior.
How do people think that distribution will change now that supply has increased (at least) tenfold? If the consensus on X.com is to be believed, it's all about personal brand. This used to be the case, but perhaps 80% of it. Is conventional discovery dead now?
[0] - If this includes organisations, like publishing houses, then fair enough. But realistically we should be able to distinguish actual authors.
I interpret this more as an underlying social contract: Both writing a book and reading it used to be laborious. Now they're not.
Reaching out to an author about specific details used to mean you've deeply engaged with the work. Writing a book used to mean you spent lots of time understanding and writing about the topic.
Neither is now true, so it makes it harder to distinguish what's worth paying attention to.
I've definitely seen this happen with myself. When I shop for books, I do extra research into the author and publisher for anything published after 2022.
That's a huge opportunity for you.
Learn the many techniques the industry has developed for scaling verification.
I don't mean infra; CI is solved. I mean Red-green & test-driven development, the testing pyramid (most unit tests/fewer functional/integration tests/happy-path e2e tests), learn how to get the most out of tools like Playwright and test frameworks like pytest and vitest and go and rust's testing tooling. Learn when to mock and when to include your code's dependencies in a test (test boundaries). Learn patterns like dependency injection so your tests can have no I/O and be FAST.
Learn to scale verification, and you'll solve your team's biggest bottleneck.
- Be able to analyze a situation into processes and discrete steps
- Be able to describe the actual (!) problem you want to be solved
- Be able to describe how to verify that the problem is actually solved
- Be able to ask the right questions regarding HOW the problem has been solved in order to gain understanding and to identify the next steps and improvements
- Be able to identify when the AI gets stuck or down a rabbit hole and to guide it out of there.
- Understand what LLMs are good at and what they are not good at. And there are tons of things they are not good at.
Disclaimer: This is absoluty not based on any scientific study or so :-) Just my personal experience! Feel free to disagree.
But my experience comes directly from my work: I have a lot of customers on my platform that are not engineers, but make use of AI to solve business problems through automations. Nearly everytime when they are stuck (or burning through a sh*tload of tokens in no time), it is because one of those skills aren't present. So I try to teach them more and more how to "think like an engineer". And often when it lands, they become much better at using AI in a very short time. The best learners are often those that already had some kind of engineering background.
Now granted, the frontier models can help themselves pretty well and often can produce amazing results with very bad input. But they are also expensive. If you work with less-than-frontier, those skill requirements become very salient again. And thats what we do. We mostly work with models like Deepseek flash or GLM Flash or even local ones like Qwen3.7 .
To successfully steer AI, you need essentially the same stuff you'd need if you were to steer humans in a management role. Meaning communication, introspection, remote introspection, emotional intelligence and all that kind of stuff.
Reading when someone is bullshitting you. Reading when someone is insecure and overcompensates that. Reading the actual problems people struggle with and not the ones they claim they do nor their suggested solutions.
Verbalizing what you need someone to do in a way that actually enables them to do it. Knowing what you want to do to a level where that is possible
Of course you will. But that's not the point. The point is that Llama runs just about every model on just about every hardware. You can always improve on any metric by dropping that constraint. But there are a lot of such metrics:
- ease of use
- range of hardware
- accuracy of results
- range of models
Probably others. Llama.cpp may not be best at any one of those. But across all of them it has been consistently good. They've shown they're willing to stay the course as well. This is a foundation level project much like Linux or Apache.
> Minimal
> We like simplicity and aim to keep the codebase as small and as simple as possible
I'm glad their philosophy extends to the website as well. Genuinely refreshing.
edit: Just to verify: I had planned to upgrade today, ran my usual benchmark prior and after and the runtime went from 4:22 to 3:59. That's a really long prompt (50K tokens) so prefill is a good chunk of that and the rest is spent on token generation.
For example, DJ Delorie's djgpp project, built in 1989, was how I got into "real" programming on MS DOS.
Wishing you unlimited power and may you never see a day of low battery. Respect and Love. Keep it up.
We also updated https://huggingface.co/unsloth/DeepSeek-V4.1-Flash-1017-GGUF with UD3.1 quants
And also https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF as well!
We provide our benchmarking methods like what graders we use, infra fixes and issues and more in https://unsloth.ai/docs/basics/dynamic-3.0-ggufs/dynamic-3.1...
Your feelings here have no technical basis. TDM is pre-provisioned, bandwith-reserved, latency-capped network that was probably hard for packet switched network 25 years ago. But with all that has gone in -- switched ethernet, qos, oam -- there is absolutely no reason to go back. Heck, even whatsapp calls across 12-hour time zones are quite crisp and it is free.
For industries that serve customer bases that are down-cycled on weekends, the dominant effect is "Deploy on Friday to minimize customer disruption chance."
For industries that operate with a 24/7 uptime expectation, the dominant effect is "do not deploy on Friday so you don't have to pay for your expensive software engineers pulling extra hours on a weekend. Much better to have disruption while the folks most qualified to fix it are in-house and alert."
There's an unverified mention of a fiber cut, which might result in fewer circuits being available. That usually returns an "all circuits busy" signal.
None of the explanations will make sense until they stop coming from unqualified political appointees.
Of particular note is this section: Risks of FTI Outages for Existing Operational Services As of December 2009, Harris reported that there are a total of 20,982 services operating on the FTI network. While FAA has transitioned all 1,808 IP services supporting FAA flight plans to the new FTI fiber optic network, additional risks will be introduced when the Agency begins transitioning other FAA services, such as En Route Automation Modernization (ERAM) and NextGen platforms, which also require IP technology to operate. Moreover, there are 19,174 existing services that may be transitioned to the new FTI fiber optic network. According to FAA, transitioning these remaining services is important because FTI now connects older FAA systems that provide safety-critical voice and surveillance radar information by utilizing Time Division Multiplex (TDM) technology, instead of IP services.
I think the problem with a lot of the push to "modernize" is we forget the legacy system has been real-world tested for 40 years now. It's the new stuff that needs to prove itself in terms of stability.
Sec Mullin: "What it appears to be right now is, as Sec Duffy and FAA was going in and putting patches into the system, they were putting new equipment and new technology. And that new equipment and new technology, it was overwhelming the older system. Which is why Sec Duffy made it very clear, the patchwork doesn't work. We need one complete system."
Speculation on my part, they just rolled out their SMART air traffic routing system. https://www.flyingmag.com/faa-smart-ai-now-active-at-22-artc...
Bet the increased strain on data polling from this system is the culprit for increased traffic.
The press conference was started 10+ min late and the first 15 min was spent on blaming Congress instead of giving an update. Kinda frustrating for those who are affected and waiting to find out what they should be doing at the airport. The Q&A has been just an endless repeat of the same few lines of blame shifting. It’d be nice if they could actually give actionable updates.
https://www.nytimes.com/live/2026/11/13/us/ground-stop-natio...
I wouldn’t say anything but this is approximately the only time in my entire life this scene will be relevant.
I worked for a person who had a fixation on "neat" deadines: the "first of the month", "the end of the week", "the start of the quarter", etc. It'd be one thing if the dates were tied to legal compliance (tax filings, etc), but that usually wasn't the case. They were chosen because they were "tidy".
I'd time meetings looking to maximize time for work. You try to meet late in the day Wednesday because an early Tuesday meeting might get an "end of the week" deadline. A late Wednesday meeting was more likely to garner a "end of next week" deadline. Meeting Thursday or Friday guaranteed an "end of next week" but you lost Thursday and Friday work time. Wednesday was a gamble, though, because it was still early enough to get you an "end of the week". Meeting after lunch helped with that.
You always had to be aware of the numeric calendar day, though. Gunning for an "end of next week" deadline when next Wednesday was the end of the month (or worse, end of the quarter) almost guaranteed you an earlier deadline than you wanted.
All of this deadline-setting happened with little regard for the amount of time work was going to take.
It was immensely fun, when setting a deadline in this person's absence, to get a group to agree on a mid-week deadline. You could see it make this person's skin crawl hearing that the work would be done on Tuesday or a Thursday.
(They also own a small number of private VHF stations for relaying company messages to air carrier operations domestically, but these aren't used for ATC purposes.)
Otherwise, the VHF radio links are operated directly by FAA TechOps.
L3Harris owns the telecommunications backhaul between facilities.
And that's why you never deploy to prod on a Friday.
https://www.wsj.com/us-news/faa-temporarily-halts-flights-ac...
> The outage appears to be affecting a telecommunications network overseen by L3Harris Technologies, according to people briefed on the matter. The network handles crucial radio frequencies that air-traffic controllers use to communicate with pilots, the people said. It wasn’t immediately clear whether the outage also affected the radar feeds that controllers use to keep aircraft safely separated in the skies.
https://www.cbsnews.com/news/faa-issues-nationwide-ground-st...
> An update to the private communications network that connects America's air traffic control system caused the system to reboot, leading to a crash and a communications outage, multiple sources told CBS News. The system is operated by L3Harris, which did not immediately respond to CBS News' request for comment.
That's just a scenario off the top of my head, it could be a lot of stuff. Plugging something new into an old, complicated system isn't risk free.
Probably someone prompted an agent: solve global CO2 emissions :-)
"The FAA in June awarded Boston-based startup Air Space Intelligence a contract — worth up to $875.9 million — for an AI-powered tool to help air traffic controllers manage the nation’s skies and a related system for an FAA nerve center in northern Virginia to better handle plane congestion. The AI platform’s initial launch in the Washington area has sparked concern among Democrats in Congress who say it should be put on pause until they have a better understanding of how it’s being deployed."
"Over two dozen lawmakers have called on the FAA to pause its 90-day evaluation period for the platform, which began late last month in the Washington region. They say there’s been an “egregious lack of transparency.”"
https://www.politico.com/news/2026/10/15/faa-stake-ai-compan...
Feels like somebody was pushing their preferred company for a government contract and now is push for the government to invest.
Seems like there are three plausible causes right now given the extremely limited communication from the govt.
1. The AI system rolled out yesterday had more access than publicly claimed and managed to break things.
2. The ongoing large scale equipment upgrade went wrong.
3. General incompetence from a department run by an unqualified realty show actor, within an administration that is actively hostile to expertise and experience in the federal workforce.
If you don't take full advantage of that, you're missing out on possibly your only opportunity ever to fix whatever's needed fixing for years.
> The FAA has reassured the public and airlines that SMART only suggests routes and doesn't affect air safety.
You'd say "Valid from the thirteenth at two zero five three zulu" over radio.
> The agency wrote on Nov. 10 that after the D.C. test, it will “now use SMART in the airspace of all 22 centers across the country to reduce delays, reduce fuel consumption and improve the travel experience for the flying public.” The tech was already being deployed at some airports.
https://gizmodo.com/the-faa-is-rolling-out-its-ai-air-traffi...
Looking at LAX departures, it looks like nothing has departed since about 10:30 AM PST, and everything later is showing Delayed or Cancelled. There are times after Delayed, but it's not clear those times mean anything. Some are now in the past.
"flysfo.com" is too overloaded to display departures. Stuck at "loading...".
JFK departure status shows almost solid "Delayed".
No good explanations anywhere.
> “This isn’t going to just be be resolved and turned around within the hour"
https://www.nytimes.com/live/2026/11/13/us/ground-stop-natio...
https://globe.adsbexchange.com/?icao=a519b8
It's Delta Airlines flight 1070
https://www.flightaware.com/live/flight/DAL1070
Edit: It landed safely. Continuing to watch our airspace clear out, most traffic at 40,000+ feet is inbound international flights
After 1962 there was too much air travel to continue to shut it down for exercises, and NORAD switched to smaller, regional level things that did not shut down air travel after that. But the results of Sky Shield III in 1962 were the only experience that the FAA had to work from when they did the full ground stop after September 11th. Sky Shield III had tested how quickly air traffic control could clear the sky in the event of a nuclear war- the USAF had put up several hundred T-33 trainers, flying them randomly around the US and Canada, and then at a certain point in the exercise the ATC had to get them all on the ground far away from major metropolitan centers (presumed lost in the war).
https://spectrum.ieee.org/lost-radio-contact-leaves-pilots-o...
(I was near ORD at the time, flying toward the outage area. We diverted to ORD, and the pilot was very friendly about it.)
At the time I thought it was Microsoft's fault, as Windows 95 and 98 were susceptible to crashing when GetTickCount() overflowed, and this ATC crash was due to a 32-bit millisecond counter on a Windows machine overflowing, but in this particular instance the system was apparently not running on a version of Windows with this particular bug and the failure was just an application error.
Aren't computers great? :)
> "The government has been working to overhaul the outdated technology that air traffic controllers rely on since Congress approved $12.5 billion for the project last year."
> "Duffy has said that another $30 billion will be needed to complete the work, and the White House has begun lobbying Congress to approve it."
[1] https://apnews.com/article/flights-halted-airports-faa-f81b0...
https://www.nbcdfw.com/news/tech/a-calamity-disrupting-hundr...
[0]https://gizmodo.com/the-faa-is-rolling-out-its-ai-air-traffi...
Update: Starship launched on schedule.
ATCSCC Advisory
ATCSCC 111 DCC 11/13/2026 NATIONAL GROUND STOP
EVENT TIME: 13/2053 - 14/0400
DUE TO EXTRAORDINARY CIRCUMSTANCES AND FOR REASONS OF SAFETY.
ATTENTION ALL AIRCRAFT OPERATORS, BY ORDER OF THE FEDERAL AVIATION
COMMAND CENTER ALL AIRPORTS ARE GROUND STOPPED. ALL DEPARTING
TRAFFIC SHOULD REMAIN GROUND STOPPED UNTIL FURTHER NOTICE. INBOUND
INTERNATIONAL FLIGHTS WILL BE MANAGED ON A CASE BY CASE BASIS.edit: https://www.fly.faa.gov/ois/oisedit/adv_oisedit_dis?number=1...
I'm sure there is a reason for it but can they not be a little bit more specific?
I know you might respond "we know they have done iffy shit before, therefore worst interpretation is warranted here."
But, well, as I said - we can do that line of reasoning about anything, but our discussions kind of break down at that point - we're just throwing down impressions and "could be's".
Don't over-index on this choice of words. LLMs can just as easily generate a diff that removes or refactors code, they don't have any intrinsic "preference" to always add more code, unless it's been trained into them.
A lot of the early training had concentrated on "wow factor" one-shot style demoable use cases, where the models indeed ended up with a tendency to assume they were in a "produce something new for me right now" situation and often or always started churning out additional code, even without all relevant information. As training regimes improve, and as customer demand evolves, this is changing, and the latest big models place a lot more emphasis on maintenance tasks - still not enough, but it's at least in the mix. As they mature, I'd expect the models' "instinct" to gradually align with what everyone's idea of a senior software engineer is, taking time to figure out if they can simplify and refactor rather than always adding more.
It's a new field, it will take time to get it right, but I don't doubt it will eventually be right.
> Because they are trained on mediocre code that is the vast majority of what's available on the internet
Again, this is outdated. It's true that the earliest training took this "spray and pray" approach, but data quality filtering is now a huge focus and the labs have made a lot of effort (and progress) in figuring out what good code is and getting the models to follow that. They've massively improved, in my experience, and it'll only get better. Again - give it time, this is new.
> "We see individual elephants leaving and possibly dying at accelerating rates, but new individuals are not coming into the population,” said Madsen. “With fewer individuals in the population, those who are left will naturally have fewer relationships. However, the remaining elephants are also changing their relationships with each other."
In my opinion "dying at accelerating rates" is the/a key takeaway. Notably climate change doesn't seem to be explicitly mentioned and hate to see poaching be a direct cause:
> Many elephant populations are already impacted through deaths resulting from poaching and conflicts with humans. Other contributing factors may be disruptions in habitat connectivity and less food available...
Other comments mention other intelligent and social species and this reminds me of Puget Sound Southern resident killer whales [0] and their decrease in food availability and offspring viability.
[0]: https://www.nature.org/en-us/about-us/where-we-work/united-s...
As for world over production collapse, Saudi has got enough capacity. Even we (India) have good amount of spare refining capacity available.
The current fires are on a completely new level that make humans ants as the fire reaches into the sky. No fire fighting force even from those from countries like mine would have a chance, and we Australians know when fires are that scale theres nothing we can do but run. These fires are like the bush fires we have seen.
Lets hope Ukraine can continue to attack all other refineries because i hope theres nothing stopping them.
Obviously Russia doesn't publish any data about how long it takes them to repair their refineries, but some of them showed significant damages from the pictures and videos you can find online.
One day, Ron (a hardware engineer from Xerox) gave a talk about the theoretical performance of an Ethernet file transfer. He calculated the end-to-end latency and data-rate including disk latency, network latency, the number of clock cycles necessary to send and receive the data, etc.
He had (overhead projector) slides that showed all his assumptions and calculations. Finally he displayed a slide with a graph showing this theoretical maximum, the actual performance of the 3Com network stack, and (IIRC) the Novell stack. As I recall, actual performance was about an order-of-magnitude lower than the theoretical max.
His point wasn't that our software division should be able to achieve the theoretical max. It was that if we (for example) got halfway there, we would know that Novell would never ship software that was twice as fast as ours.
He was very nice and scholarly about it and did not mean offense. I and many others weren't offended. But a surprising number of people were.
I'm sorry to hear that he passed.
ep0: 3c509 in test mode. Erase pencil mark! This means that someone has
scribbled with pencil in the test area on the card. Erase the pencil
mark and reboot. (This is not a joke).
It's the E1 pad between the barcode and the empty BIOS socket: https://en.wikipedia.org/wiki/3Com_3c509#/media/File:3Com_3C...Details (pg 88): https://www.ardent-tool.com/NIC/3c5x9b_Technical_Reference.p...
----------------------------------------
It is the user’s responsibility (with help from the configuration program) to avoid configuring the boot PROM on the adapter in such a way that the system is not able to boot. If this does occur, you can manually jump the Test Via using a #2 pencil. The Test Via forces the adapter into test mode, which disables the boot PROM so that the adapter configuration program can be run. You must cover the designated area thoroughly with the pencil mark.
Test mode forces the adapter not to perform the Automatic Initialization sequence, which means the EEPROM is not read and the adapter is left in the following configuration:
■ Address Configuration register = 0000h
■ Resource Configuration register = 0000h
■ Product ID register = 0000h
Test mode also shortens the ID sequence to 8 bytes (first is still FFh and last is 69h) and forces it to the active state (that is, the adapter is active and will respond to I/O cycles at base address 0200h even without going through the ID sequence).
After you are done, thoroughly erase the pencil mark.
----------------------------------------
Ron Crane which was 3com's 4th employee and designed the 3C100, 3com's first ethernet product. Many years later he hired me for my first real job of out school and brought me to the bay area. I then got to work side by side with him and David Boggs (https://en.wikipedia.org/wiki/David_Boggs). At the time I never realized how special working with these two were...wasn't every company in the bay area like this?
I stayed in tough with Ron Crane until his death in 2017. He was both brilliant and incredibly sweet. He was so proud to take me behind the scenes at the Computer History Museum to show some work he'd been going.
I know Ron has writen about 3com...but this story captures the Ron I know: https://www.theregister.com/on-prem/2013/05/28/how-god-and-u...
I was on the Macintosh network os client team. For whatever reason (maybe he was annoyed that Apple copied so much from Xerox) Metcalfe didn't like Macs. In this era Macs used Apple's proprietary AppleTalk/LocalTalk network. We did a demo for 3Com execs using a large external monitor connected via a single black coax cable that looked just like thin-ethernet. He got excited for a moment and came right over to look at the Mac and was very disappointed to see the cable attached to a display! (3Com was working on Ethernet cards for Mac at the time but they weren't ready at the time of this demo.)
For those too young to remember: there was an SF Bay Area baseball park known as Candlestick Park that was briefly branded "3Com Park" -- to the great annoyance of SF Giants fans. During one interview he as asked if in the early days of 3Com he would have done anything differently. He said, "I would have named the company 'Candlestick'!"
They were high-quality cards and the lifetime warranty proved to be very popular.
For those with textual proclivities: https://archive.computerhistory.org/resources/text/Oral_Hist...
Metcalfe has a reputation for being brash and outspoken and it clearly shows in the interview. I think it served him well in business and later in life as a pundit. I remember looking forward to his columns in InfoWorld, though I didn't necessarily give much stock to his predictions.
He talks a little bit about the downfall of 3Com in the interview, though he begs-off of talking about things that happened there after he left.
I very much enjoyed the trivia and minutiae about the creation of Ethernet and the Digital / Intel / Xerox relationship that ultimately resulted in it being standardized.
Basically, it was a long listen but there was some good stuff in there.
Sec Mullin: "What it appears to be right now is, as Sec Duffy and FAA was going in and putting patches into the system, they were putting new equipment and new technology. And that new equipment and new technology, it was overwhelming the older system. Which is why Sec Duffy made it very clear, the patchwork doesn't work. We need one complete system."
The motivation for building the new coordinator is: as people get more comfortable using Claude Code, we noticed they start running more agents in parallel — Claude works, so you start another thread to work on the next thing. As the model gets smarter and more token-efficient, and as features like auto mode and cloud let agents run longer, this has become more and more common. You can do a lot of work in parallel this way. But the longer you multi-Claude this way, the more it feels like the UX is not good enough: it is hard to manage the complexity and context switch. It feels bad to fragment your attention like that.
At the same time, sometimes you want to deeply focus on a single task, and go back and forth on it without switching your attention to something else. Not every kind of work needs parallelization. This kind of single threaded focus remains really important.
The solution we came up with is coordinators. You do your work in Projects, then talk to your project’s coordinator, which then delegates work to threads when needed. Threads run in parallel and bubble up information and questions to the main thread. You can also jump into a thread to do deep work.
If you’ve tried it, we’d love to hear what you think. If you haven’t, give it a spin and please share feedback on what we can do to make it better for your use case (run /bug or make a GitHub issue).
Liked it so much and thought I'd clean it up a bit, deploy, and share :) Enjoy!
contrary to popular belief, DNS doesn't actually need 24 hours (or whatever) to "propagate". It only needs enough time for caches to expire, if they exist. For non-existent domains this shouldn't be an issue.
Without planes, lots of them probably just tried to rush to Amtrak and they received a volume they’ve never seen in a very short order.
I doubt there is anything more complicated than that.
Quite a bit was changed between "illumos bhyve" and FreeBSD upstream. We in illumos decided to take a slightly different approach in how we approached the machinery for saving/restoring the state of devices emulated in-kernel, which is a necessary part of doing such a live migration. FreeBSD was first to add that functionality, but we (I) felt it exposed too much in the way of implementation details from that device emulation, rather than a more abstract model of the emulated hardware. It was at that point that illumos "soft-forked" the kernel VMM component so we could take a different approach. While the two kernel VMM components are very similar, and much of the bhyve userspace emulation (a counterpart to propolis) will run on either OS, there are some key differences around instruction emulation and those kernel-emulated devices which prevent them from being drop-in replacements for each other.
All that said, I'm happy to see FreeBSD making progress on this front. They've been a friendly peer in this space, and both projects benefit from efforts in this area.
1. Log in to OKTA and choose my specific account (we have 2 accounts for some reason, and only one of them is used for development) 2. Log into AWS with that account 3. Select another sub-account in AWS 4. Navigate to my developer EC2 5. Start the instance 6. In VS Code, open a terminal and log in with “aws login —profile myName” 7. Wait for redirect to AWS in the browser window 8. Click ok and close the tab 9. Back in VS Code, open remote explorer and click on a preset that does an ssh tunnel to my workspace on the EC2 10. Wait for VS Code to spawn a new window 11. Now I can go through all of my other development setup tasks (another 6-10 steps depending on what I need to do)
Every day. And our instances force auto-shutdown at 9pm, so if you need to work late, better save your work, log off preemptively, and go through the same process.
It was a solved problem long before smart phones.
Technology is supposed to improve the experience not make it worse.
1. Tap "login", taking me to the login page.
2. Tap the "login" button on that page.
3. Sign in using 2FA. (A few more taps, but I'm not counting that against the portal).
4. Tap "Redeem Parking Permit"
5. Tap "Redeem use of Permit Pack"
6. Tap the car that I used that day (which is always the same car)
7. Tap "Confirm"
8. Tap the calendar icon to open the date chooser.
9. Tap on today's date (why isn't this the default?!)
10. Tap "Confirm"
11. Tap "Complete"
After a month or so at being annoyed by all of this unnecessary friction, I had Claude write an iOS app that keeps me logged into the parking portal. Now paying for parking requires just a single tap in my custom app. It also shows me a widget, telling me whether I have already paid or not.
I've built a lot with Claude this year, but that simple tool might be the one that brings me the most happiness.
iPads especially, they aren't a tool for productive work. Not many people use iPads professionally - they are designed for consuming. If I was in school today I think the distractions they provide would make me a far worse student.
per NWS on X.
Multiple other Weather Watchers reporting that there was a single page from an office that had a cached error message that was 2 years old.
Looks like the woodpecker has arrived.
Edit: OMG I just realized who you are. I used to read your blog obsessively before I became a pilot, and while I was working heavily on mysql. Thanks for that!!
I'll message you.
This is an ETTUS B210 attached to a dual 4090 box. It's a polyphase channelizer that takes the entire airband, channelizes it using cuda code on GPU, demodulates it, detects transmissions, throws away data tx and then feeds the tx through a voice model I created for transcription. A separate LLM reads the transcriptions to create a situational awareness picture of the airspace.
I'm interested in open sourcing this and making it available for folks who can run an ETTUS B210 or high quality SDR with enoguh bandwidth and a local GPU machine, so that we can cover more airpsace.
This was born out of an incident we had not too long ago where a pilot around KORS (based at Friday Harbor) lost the engine in his 172, and put it in the water in front of a ferry and was picked up by the ferry. We're a pretty close bunch on Orcas Island and we only heard about it days later via social media. That felt completely unacceptable, and I set out to create a radio that could monitor everything in real-time, all the time, and alert on emergencies. That has extended to transcribing and creating situational awareness for a chunk of airspace.
I'm an IFR pilot. I fly a T206H regularly in IMC around here and I'm a ground instructor. You can email me at mark@defiant.com or find me on x at mmaunder if you'd like to connect. I'm particularly interested in hearing from folks who can run a station on their own hardware and are under bravo or a charlie or even just somewhere interesting re airspace.
A spokeswoman for the National Weather Service said a screenshot circulating online that claimed its weather radar was down was “outdated.”
The Weather Service “has not experienced a disruption to its radar systems today,” said Kim Doster, the spokeswoman.
ATCSCC Advisory
ATCSCC 111 DCC 11/13/2026 NATIONAL GROUND STOP
EVENT TIME: 13/2053 - 14/0400
DUE TO EXTRAORDINARY CIRCUMSTANCES AND FOR REASONS OF SAFETY.
ATTENTION ALL AIRCRAFT OPERATORS, BY ORDER OF THE FEDERAL AVIATION
COMMAND CENTER ALL AIRPORTS ARE GROUND STOPPED. ALL DEPARTING
TRAFFIC SHOULD REMAIN GROUND STOPPED UNTIL FURTHER NOTICE. INBOUND
INTERNATIONAL FLIGHTS WILL BE MANAGED ON A CASE BY CASE BASISAny evidence for these assumed causal relationships? Wars have been waged by small numbers of people in the past and today, migration in such general terms has always been inherent to humans, resources are depleted by mass-consumption of a tiny fraction of the global population, and hunger is caused by not distributing available food to those who need it.
These problems are hence mostly political. Again, only a tiny fraction of the global population has any significance in the relevant decisions.
I try to understand this worrying mindset but it still seems like short-term thinking. Essentially all of our problems are caused by excessive population - climate change is caused by energy usage which is too high because there are too many people. Wars, migration, resource depletion, hunger, water scarcity, all of it. So we (as in, human civilization) short-term think our way into rationalizing that it'll be ok because technology will provide a solution and we don't need to make any real changes.
Now the population trend looks like it's about to reverse and we can finally see a way to a possible future without scarcity, and we short-term think that into being a problem because we might find it difficult to afford to care for our elderly anymore. But we can! The economic engine will never be more powerful than it is now: so exploit it, use it to solve the pain of population contraction while we adjust our economic and productive structures so they can serve us effectively in a world where growth isn't purely consumptive.
I'm sure this is all hopelessly naive.
If you haven't watched it, Soylent Green with Edward G Robinson and Charlton Heston is an illustration of this (no spoilers please!) It was based on Harry Harrison's novel Make Room, Make Room where population growth was out of control, and human life was worth so little that the police used mechanical scoops to clear rioters from the streets.
And then on the other side, we have Blade Runner where Earth's population is declining because of off-world colonization and replicants are created for labor.
If people's MS subscription revenue were funneled into on-site devs and devops for open source solutions, we'd all live in a better world. Sure, the money wouldn't be funneled to a single ledger in Redmond, but the overall productivity would be higher and there'd be greater GDP growth in other areas.
I've heard of so many small companies with a slightly technical bent replacing their SalesForce with better CRMs that make everyone happier, thanks to AI. MS is next.
Let's make sure that dollars get funneled away from economic rent and into more productive efforts. Schumpeterian rents might exist for a bit, but they should never exist long. Microsoft's rents have lasted for too long to be economically optimal.
> What shipped, who's using it, and how it's helping them matters FAR MORE than lines of code, prs merged, tokens spent, or the number of agents running in parallel.
Sometimes when I hear shocking amounts of money spent on AI tokens, I wonder if someone's obsessing over the tech and forgetting that only the result matters.
The manufacturers of drugs are highly regulated.
The prescribers of drugs are highly regulated.
The disbursers of prescription drugs are highly regulated.
The shippers; not so much.
There's a good reason the mob likes logistics companies.
https://en.wikipedia.org/wiki/Toronto_Pearson_International_...
The scenario you outlined is not all that common. There are much more sophisticated scams. What I see the most often are disclosures of load information and pickup numbers to bad actors (e.g. via social engineering).
When you roll up to a facility with a truck, you need to provide information from the bill of lading, and the key piece of data is the pickup number, which is basically like a PIN for the shipment. If that gets into the wrong hands, then someone else can show up at that facility, pose as the carrier, and get loaded with the goods. It's quite low risk during the operation because there's no hijacking or anything else obviously criminal happening. People will literally load the thief's truck.
While probably harder to find a no-questions asked buyer, the price premium on GPUs is a lot higher than some consoles.
Commercial shipments don’t work like when you receive a package from Amazon. The terms of the freight, including the point at which ownership is handed off, are negotiated terms. An $11.4 million contract would have this negotiated and agreed upon. Depending on how it’s structured, the shipment may not have been the shipper’s responsibility at all after it left the warehouse.
> I can't imagine their insurance is too happy about it either.
Insurance would take the risk of the transport method into account. The policy would say what transport methods are covered and the allowed methods would be used to calculate the risk of loss.
Tech-Insider $11.4M of Nvidia Chips Vanished in Oakland (6 points, 1 week ago) https://news.ycombinator.com/item?id=50394165
TomsHardware Six pallets of Nvidia GPUs worth $11.4M stolen in California in Aug 2026 (7 points, 1 week ago) https://news.ycombinator.com/item?id=50394256
ArsTechnica Nvidia hardware worth $11.4M vanished between Oakland and Texas (9 points, 6 days ago) https://news.ycombinator.com/item?id=50405863
TechPowerup (10 points, 6 days ago) https://news.ycombinator.com/item?id=50407874
— Charles Eames