highlights | Hacker News (2024)

I worked on this experiment as an undergrad ~10 years ago during my freshman year! We built a Cherenkov radiation detector, focusing magnets, and did tons of simulations.

This is all from memory, but I remember the beamline setup was to get protons from the accelerator there, smash them into a target, which produced various charged particles which could be focused with the magnets, sent down a long pipe where they would decay into neutrinos et al. Then, there's a near detector and a far detector (far detector deep underground in South Dakota). The aim is to measure the neutrino flavors at both detectors to better understand the flavor oscillations (and look for asymmetries between neutrino/anti-neutrino oscillations, hopefully to help explain the matter/antimatter asymmetry in the universe).

The particular bit I worked most on was studying the effects of adding an additional solid absorber at the end of the beamline, which was needed to absorb all the particles that didn't decay in the pipe. It would produce more neutrinos that were unfocused, so would affect the near-far flavor statistics (since these would be detected at the near detector but not the far since they were unfocused, ruining the statistics). It was a great intro to doing physics research :-)

I worked on this movie, I was at DNEG at the time. One of the standout things that I remember is that this particular simulation was toxic to the fileserver that it was being stored on.

From what I recall, I don't think that it was running on that many machines at once. Mainly because it required the high memory nodes that were expensive. I think it was only running on ~10 possibly 50 machines concurrently. But I could be wrong.

What it did have was at least one dedicated fileserver though. Each of the file servers at the time were some dual proc dell 1u thing with as much ram as you could stuff in them at the time (384 gigs I think). They were attached by SAS to a single 60 drive 4u raid array. (Dell PowerVault MD3460 or something along those lines. They are rebadged by Dell and were the first practical hotswap enclosure that took normal 3.5" SAS drives, that didn't cost the earth)

The array was formatted into 4 raid6 groups, and LVM'd together on the server. it was then shared out by NFS over bonded 10gig links.

Anyway. That simulation totally fucked the disks in the array. By the time it finished (I think it was a 2 week run time) it had eaten something like 14 hard drives. Every time a new disk was inserted, another would start to fail. It was so close to fucking up the whole time.

I had thought that the simulation was a plugin for houdini, or one of the other fluid simulation engines we had kicking around, rather than a custom 40k C++ program.

This was such important and transformational work and I remember at the time being quite dismissive of it.

I knew Orr’s and Suchman’s work (they worked in a physically adjacent area, but completely different group, though we were all under John Seely Brown and because they were nice people). Thankfully I was grown up enough to be polite, but really I was such a techno-determinist that I figured user problems came from ignorance.*

To be fair, I was not the only one: the insights described in this book draft surprised a lot of people, not just how they improved the copiers but how those two even approached the problem (starting with the sociology of the repair workers). It sure surprised Xerox management. But I’ve heard it said many times that this work led to restructuring the paper path in a way that justified (paid for) everything spent on PARC.

I did grow up of course and now do see my work (machines, chemistry, etc) as a small part of a large social system. A successful company has to base its product plans starting this way.

To choose an example of failure to appreciate the social scope (but not pick on it) the crypto folks spend their time on technology, based on a social model they want to exist rather than the one that currently does. I think it’s a big reason why it’s barely impacted the world in, what, 15 years? Xerox was the same, and it helped them sell a lot of copiers, but didn’t make them as ubiquitous as they could have been. Another example: everybody laughs at Google for launching “products” that go nowhere and are quickly forgotten. We all know it’s because of a screwed-up, internally-focused culture. But sometimes a product succeeds without marketing (e.g. gmail, at the time) because it happened to be matched to the actual, external need. It makes this kind of continuous failure even more damning.

* TBH, 40 years later I have not 100% shed this view — e.g. my attitude towards complaints about git. Maybe this means I’m still a jerk.

I'd forgotten how close to printing machines the old photocopiers were. You would basically either have a crap one you could operate yourself, or take your stuff to the printery to have professionals (a subset of librarians I think, or the logical join over librarians and computer operations staff) do it for you. Printing machines had a fleet of maintainers, craft unions who walked off the job if you touched a dial.

They were amazing at doing things which really mattered: shrinking an A0 architectural drawing down but maintaining aspect ratio. Adjusting offsets for the print for binding signatures, so the 1st and 16th page was not too far out because of wrapping around the other 8 pairs of pages. Even just working out how to rotate the pages for N-up printing. But the GUI sucked. I think they called ours "the bindery" because it's main gig was doing PHD from soup to nuts, binding included.

The repair techs had the most amazing flight cases, packed with tools which served one specific purpose.Like, A doohickey to adjust the corona wire, without dismantling the imaging and toner roller, with a tonne of equipment hovering over your head on a gas-lift. Screwdrivers with very very carefully chosen lengths. Torque wrenches. It was high tech meets motor racing meets.. IBM.

I am told they were paid better than many computer techs. The IBM guy was paid IBM scale to fix it on IBMs timescales. the xerox guy did more random shit, with more devices, more often.

They had a very corporate look. that amazing briefcase or six. Suit, tie. Very acceptable.

I know a guy who worked for a paper-folding-and-envelope-stuffing company and it was very similar culturally: can-do, fix anything, but working on giant multi-million dollar machines which were used twice a year to do tax mailouts, and election materials, and the rest of the time rented to the original spam merchants for 10c per thousand mailouts. The secondhand value of these machines were like photocopiers: Really significant. He was brought out of retirement to help take one apart into TEU equivalent chunks to be shipped to Singapore from Brisbane. His retirement gig at one point was repairing Espresso machines, he said it made him feel familiar and useful.

The era which was the end of the typing pool was fascinating. All kinds of arcane roles which only make sense in the absence of email and tiny printers everywhere. Some of those jobs had been there from the days of hand-copying, Dickens-era and before.

I was a college swimmer, qualified for Olympic Trials in 2012 and 2016. There are absolutely slow and fast pools. It basically comes down to two things:

1. The depth - which is only 7ft in Paris, unusually shallow for a competition pool.

2. The sides. Does the water spill over the sides into the gutters, or smash into a wall and bounce back, creating more chop.

A trained eye can see all the swimmers in Paris struggling in their last 10-20 meters (heck, an untrained eye can spot some of these). Bummer that it makes the meet feel slow but at least it generally affects all the swimmers equally

I went to an engineering school, and one of the stories the old boys told was that at some point the city had built a new bridge, and tendered the destruction of the old bridge, and we'd put in the winning bid.

The scheduled day came, but only an hour or two after the scheduled time an urgent messenger came from the city: the neighbours were complaining, could they please just destroy the bridge all at once with the next explosion?

It turns out the civil engineers had been enjoying themselves in the interval, checking their modelling by seeing how many parts of the bridge they could blow off of it, while leaving the majority of the structure still standing...

“The Heavy Press Program was a Cold War-era program of the United States Air Force to build the largest forging presses and extrusion presses in the world.” This ”program began in 1944 and concluded in 1957 after construction of four forging presses and six extruders, at an overall cost of $279 million. Six of them are still in operation today, manufacturing structural parts for military and commercial aircraft” [1].

$279mm in 1957 dollars is about $3.2bn today [2]. A public cluster of GPUs provided for free to American universities, companies and non-profits might not be a bad idea.

[1] https://en.m.wikipedia.org/wiki/Heavy_Press_Program

[2] https://data.bls.gov/cgi-bin/cpicalc.pl?cost1=279&year1=1957...

Wow, this hits close to home. Doing a page fault where you can't in the kernel is exactly what I did with my very first patch I submitted after I joined the Microsoft BitLocker team in 2009. I added a check on the driver initialization path and didn't annotate the code as non-paged because frankly I didn't know at the time that the Windows kernel was paged. All my kernel development experience up to that point was with Linux, which isn't paged.

BitLocker is a storage driver, so that code turned into a circular dependency. The attempt to page in the code resulted a call to that not-yet-paged-in code.

The reason I didn't catch it with local testing was because I never tried rebooting with BitLocker enabled on my dev box when I was working on that code. For everyone on the team that did have BitLocker enabled they got the BSOD when they rebooted. Even then the "blast radius" was only the BitLocker team with about 8 devs, since local changes were qualified at the team level before they were merged up the chain.

The controls in place not only protected Windows more generally, but they even protected the majority of the Windows development group. It blows my mind that a kernel driver with the level of proliferation in industry could make it out the door apparently without even the most basic level of qualification.

Retired from Apple a few years ago (at age 57).

I was not obsolete. A big company like Apple, there are always things that need taken care of.

I assumed with iOS, Swift, etc., maybe the guys on the Cocoa team were obsolete? Of course not. That code is still there, still needs maintaining, interoperability with the new languages, frameworks, etc.

I'm more surprised they want to stay on.

And that is in fact why I left Apple: the job had changed, the "career" had changed. The engineers were no longer steering the ship. It had been that way when I started in 1995 though. A "team", let's say the graphics team, would figure out what API to revisit, what new ones to add — perhaps how to refactor the entire underlying workflow. The "tech lead" (who would regularly attend Siggraph since we're talking about the graphics team) would make the call as to what got priority. Marketing would come around after the fact to "create a narrative" around all the changes to the OS. I hate to say it, but many, those were the good ole' days.

(And let's be clear, in the 90's, Apple's customers were more or less like the engineers, we also loved the machine for the same reasons they did — so we did right by them, made changes they would like because we wanted them too. You can't say that as convincingly for the phone, being a mass consumer device.)

Marketing took the reins long ago though — especially as Apple began to succeed with the iPhone (which, someone can correct me if I am wrong, but I think was an engineer driven project initially — I mean most things were up to that point).

I stuck around nonetheless though because there was money to be made and kids still to raise.

When the last daughter flew the coop though, so did I.

I once bought a vacation home that was a century-old English cottage that went through 7 different owners over time. It once belonged to a US state senator. Another time it belonged to a prominent local businessman who went to jail for white collar crime, and went through a nasty divorce. Anyway, the house had a TL-15 Star Safe embedded in the wall in the master bedroom. The previous owner did not know the combination. Neither did the owner before him. Some unknown person at some point had attempted to open it, as the safe had 3 drill holes on the face plate.

There was a very old sticker on the safe bearing the name of the company who apparently installed it. The phone number was so old it did not have an area code. Fortunately the company still existed after multiple decades. I called them and asked if they could open it in a non-destructive way. One of their technicians came, looked at it and probed it for a couple hours, but determined he could not open it. And the combination had been changed from the manufacturer's default. He gave me the contact info for a reputed safe technician who could help.

Later I called this safe technician, but he was incredibly difficult to get a hold of. I had to leave multiple voicemails and send multiple emails. We chatted briefly one time and he said he would get back to me later to schedule an appointment. But he seemed half-retired and not interested in the job, as I never heard back, despite multiple contact attempts and my offer to pay handsomely. Eventually I became frustrated with his non-responsiveness and stopped caring about the safe.

Fast forward a few years later, I was going to sell the vacation home, but I really wanted to open the safe before selling. Curiosity had gotten to me. I searched online for another safe technician, and found a supposedly reliable guy. I arranged an appointment. He showed up a few days later. I asked him to open it any way he could, even if he had to destroy the safe. He started drilling, making multiple holes over the course of 2 hours. Eventually he came to me and said he ran out of drill bits as they all got worn out. He had to leave and promised he would be back.

It took one week for him to eventually come back early one morning with more drill bits. He spent another couple hours drilling. Then he put a camera scope in the holes and claimed he could see 3 of the 5 wheels spin while the other 2 were broken. He spent an entire day trying to manipulate the wheels. But after a whole day of work, he came to me with a defeated look and apologized saying he was sorry but he doesn't think he is able to open the safe.

I went back online to find yet another professional who could help. I learned that what I really needed to look for is a professional who is a member of SAVTA (Safe & Vault Technicians Association). So I found a SAVTA tech who on the phone told me a TL-15 safe in a residence is unusual as it is normally made for businesses like a jewelry store. Unfortunately he said his next availability would be about a month from now, and I was going to sell the house in the coming weeks.

Eventually I found another SAVTA tech who was available on a short notice. He and a colleague both arrived a morning, and it took them 3 hours to do more drilling and more manipulation to FINALLY open the safe.

Guess what was in it?

Nothing. It was empty! I closed the sale of the house literally 2 weeks later. I was still very relieved to have gone through this hassle to open it. The unsatisfied curiosity if it had not been open would have eaten me alive :) Also I decided in my next house I wanted a safe rated TL-15, as clearly they can withstand a lot.

I worked in the snowmaking industry at ski resorts for more than a decade before getting into tech. Many ski resorts have a snowmaking reservoir at elevation and a pumping system to fill it (usually off peak) and then use gravity to actually feed the snowmaking guns (at least partially). Almost every snowmaking manager (that I talked to) has had the idea at some point to try some sort of pumped hydro offset, but I'm unaware of anyone who has actually tried it. It would be fairly small scale (reservoirs can be ~20 million gallons, usually less) but it would be interesting to see the economics of it because the infrastructure is already there (pumps, pipes, reservoirs, etc). The systems generally even sit unused for 7-8 months of the year.

I think some of the challenges are that while most resorts have a fairly massive pumping system, it's usually geared towards slowly filling the reservoir, with the rest direct feeding the snow guns. Not many places have the need to fill a 20 million gallon reservoir in a couple of days.

There's also the probability that the head pressure's wouldn't work out. Gravity feeding from an upper reservoir near the top of a large mountain can result in thousands of PSI at the bottom if not passed through a series of pressure relief valves. I'd imagine ideally you would have to build a generating station and a new catch reservoir at the perfect elevation because if you are pumping a lot higher than needed the efficiency is going to drop significantly.

Based on my understanding, some of the details he gave about the Spyglass/Microsoft situation are not quite right, but I don't think it would appropriate for me to provide specific corrections.

However, since I was the Project Lead for the Spyglass browser team, there is one correction I can offer: We licensed the Mosaic code, but we never used any of it. Spyglass Mosaic was written from scratch.

In big picture terms, Marc's recollections look essentially correct, and he even shared a couple of credible-looking tidbits that I didn't know.

It was a crazy time. Netscape beat us, but I remember my boss observing that we beat everyone who didn't outspend us by a favor of five. I didn't get mega-rich or mega-famous like Marc (deservedly) did, but I learned a lot, and I remain thankful to have been involved in the story.

I remember being underwhelmed by the www before the graphical browser. Gopher I felt was superior. I would read about the graphical web browser in magazines but it required a slipConnection which may not have existed at this point.

One day I read about a guy in brooklyn who had a website at www.soundtube.com and was selling music on the internet. I got in touch and went to his office in brooklyn to look at his website in a graphical browser.I than followed his lead in getting setup.

The logo for the site was a half squeezed tube of toothpaste with the word sound tube on it.

I don’t remember his delivery mechanism. The last time I visited the site it was the same logo but with the subtext that “what could have been”.

I occasionally look for more information about sound tube.

Seems to be lost but I hope it is only missing.

> You can choose a soothing music play list in the car and it automatically resumes in the next ride

Oh wow! There is a non zero chance that was implemented because of some feedback I provided as a trusted tester many months ago. I napped my son in them a lot when they were free and just spent my time thinking up things they should do and reporting them in app.

The best analogy of zk-proofs I've heard is to suppose you have found Waldo in "Where's Waldo," and want to prove that you have done this without revealing the location.

You could take a piece of paper (much larger than the picture/book), and cut out a waldo-shaped hole it and position the paper such that he is shown in the hole. Then, when you show it to the challenger, they know that you have found him without you revealing where he is.

So sad! Rest in peace.

I got to meet him in his office and have dinner with him once. It was an unforgettable and hugely influential experience.

Two fun anecdotes that have never left me:

- He taught me that IPIs (inter-processor interrupts) are inherently and hugely expensive. Knowing this has helped me with architectural choices more times than I can count.

- He quoted (I think from someone else) a rebuttal to the idea that Physics is the reality and math is just theory. It goes something like: Math is the reality that physicists sometimes discover. Love it.

I've actually isolated and sequenced the subject fungus (Parengyodontium album) from terrestrial sources. If you'd like to check out the photos (and DNA), check out:

https://www.inaturalist.org/observations/147456216

https://www.inaturalist.org/observations/150149352

In primary care, I used to smell sinus infections/strep as a patient walked in the room for their “sick visit” and felt confident enough to diagnose without swabbing but I still swabbed anyway to avoid antibiotics for viral infections. I’ve long left primary care for hospital and concierge medicine so now c diff stool and melena get me usually.

There was this one time in residency I had a tiny older lady from a very rural town come in with a festering breast wound. One breast was normal sized and the other was 4x size of the normal one. Almost the size of a medium watermelon. Turns out it was necrotizing breast cancer and severe cellulitis. She didn’t recall much about why it took her this long to seek help. Her kids pleaded her to get it checked out but she had a deep mistrust of healthcare workers it wasn’t until the smell was unbearable for her family that they brought her in. The whole ED smelled of rotten infected flesh and no amount of winter fresh and peppermint oil was able to help the smell. That was about 10 years ago and I still remember that day and smell.

Wasn't expecting my question to hit top of HN. I guess I'll give some context for why I asked it.

I work in quantum error correction, and was trying to collect interesting and quantitative examples of repetition codes being used implicitly in classical systems. Stuff like DRAM storing a 0 or 1 via the presence or absence of 40K electrons [1], undersea cables sending X photons per bit (don't know that one yet), some kind of number for a transistor switching (haven't even decided on the number for that one yet), etc.

A key reason quantum computing is so hard is that by default repetition makes things worse instead of better, because every repetition is another chance for an unintended measurement. So protecting a qubit tends to require special physical properties, like the energy gap of a superconductor, or complex error correction strategies like surface codes. A surface code can easily use 1000 physical qubits to store 1 logical qubit [2], and I wanted to contrast that with the sizes of implicit repetition codes used in classical computing.

1: https://web.mit.edu/rec/www/dramfaq/DRAMFAQ.html

2: https://arxiv.org/abs/1208.0928

After my little sister had her first child and realized how expensive baby stuff is, she started a lucrative side-hustle and ran with it for years. Basically, she bought baby stuff from a warehouse that got their inventory from returns at large retailers like Target and Walmart. She focused almost entirely on baby strollers, but also backyard swing sets for kids, and got it all for pennies-on-the-dollar. She became friendly with the customer service repos at the stroller manufacturers and could usually get replacement parts for free (it's a warranty replacement if the service rep says it is). She knew all the stroller model numbers and their associated various part numbers. She got really good at repairing the strollers in her garage, and then flipping them on Craigslist. Her garage looked like a baby stroller showroom. She made decent money doing it, but the best part is her "customers" (other new mothers, most of them poor) were always so happy and appreciative because of the deal they were getting. Everyone was happy.

The real secret sauce to her side-hustle was the relationship she had with the lady who managed the warehouse where she bought the baby stuff. The warehouses usually have auctions on large lots or pallets of stuff; you bid on whatever's on the pallet, you've got no choice. The lady used to let my sister come to the warehouse periodically (usually just before a big auction) and cherrypick what she wanted, which was always the baby strollers and swing sets. The side-hustle wouldn't have worked without that. (My sister (and her husband) used to flip houses, too, and I think she sold the warehouse lady a house.)

Hi HN the main (more detailed) article is here https://github.com/karpathy/llm.c/discussions/481

Happy to answer questions!

My uncle worked at Estes, the hobby rocket company, for many years. He was always a stickler about calling the black powder propellant sources "motors" and indeed older motors are labeled such. [1] He insisted they were not engines as they had no moving parts and would always correct me when I said "rocket engine." He eventually explained that rocket engines exist, but they are engines with valves and pumps and use liquid fuel (e.g. the Saturn V's F-1 engines), while solid rockets (e.g. Estes' products, or the shuttle's SRBs) are simply motors since they merely consist of burning propellant and a nozzle. Indeed the wiki pages for the F-1 and the SRB are consistent in calling the former engine and the latter motor.

However, at some point since he retired, Estes transitioned to calling them Engine/Motors [2], and now, the primary labelling Estes uses calls them Engines, though Engine/Motor is still printed on the cardboard casing itself. [3]

Interestingly, the Spanish, French and German on the motors still use motor, as Motor, Moteur-Fusee and Raketenmotor, respectively.

Because of that upbringing, I have since treated the words to mean that a motor is something that provides force of motion (thrust or rotation) - it may or may not also be an engine, as in the rocketry examples. An engine is a contraption with moving/interacting parts that uses energy to accomplish some goal - that goal may (F-1, car engine) or may not (cotton gin, search engine) be the propulsion of the contraption itself and what it's attached to.

That said, as a child I made no such distinction, hence the frequent corrections. I am happy to recognize that in common vernacular they are usually synonymous, though it would still sound strange, I think, to call something a 'search motor' (edit: however, see comment by yau8edq12i !) or a 'graphics motor' just as it would be jarring to encounter 'servoengine'.

1: https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEi...

2: https://www.apogeerockets.com/bmz_cache/f/f8ecac9604d017a5c7...

3: https://estesrockets.com/products/b6-4-engines

Former journalist here. I spent lots of time in college learning about libel law, and then applying it in my professional life as an editor.

One thing about libel that many people don't understand is that retraction and editing of the content isn't a defense. So where it says "note the libel-friendly phrasing" and "now edited to avoid any possible threats of libel" and "[editor’s note: removed a possibly incorrect claim]" he could still be found guilty of libel if previously published assertions contained non "libel-friendly" phrasing. As long as a defamatory assertion was published at some point, you can still be found guilty of libel.

It probably goes without saying, but it is also not a defense to libel to say that you asserted something to be true merely because there was no evidence to the contrary. Absent a contractual or legal obligation, Lumina had no duty to engage with him and answer his questions. So if Lumina can provide evidence that Trevor asserted things that are demonstrably false, and they damaged Lumina's business, then Trevor can't argue as a defense that he merely had no way of knowing that they were false.

Finally, Trevor seems to be saying in his update that he was merely asking questions -- but it's possible for a court to find that merely phrasing false, defamatory assertions in the form of a question is not an absolute protection against a libel claim.

It's not any kind of paradox. Structural unemployment happens when the skills of the work force don't match the needs of employers, so there is both unemployment and difficulty hiring.

Structural unemployment is usually high when there's a rapid change in demand for skills, as of course there is in tech. It results in crazy high salaries too. People with machine learning experience are getting 7-figure offers, while people with jQuery experience can't find jobs.

As an individual, you can both improve the economy AND make fat stacks by learning the skills that are in high demand. As an employer, you can do better by finding skill sets that aren't in high demand, with enough overlap with what you need that you can retrain. There are a lot of unemployed video game programmers right now, so if you can figure out how to use people with those skills you can hire some smart, energetic people at moderate salaries.