People are reacting negatively to the ads, but there's a bigger point. This is bearish as heck for AGI. If OpenAI were recursively improving their general-computer-using agent, who was going to be superhuman at every job, they wouldn't need to be messing around with things like this.
ChatGPT is a useful product, which they're monetising in a well-travelled internet company way. The bad news is you're going to have ads in your ChatGPT in 2030. The good news is you're still going to have a job in 2030.
I switched all the machines at https://lanparty.house over to Linux a couple months ago. So far, we've experienced noticeably fewer problems on Linux compared to Windows. Stability and performance are better. I can't think of one game we tried that didn't work. And wow is it nice not to have all the ads and crapware in our faces anymore.
(I'm aware that Battlefield series and League of Legends won't work due to draconian anti-cheat -- but nobody in my group cares to play those I guess.)
I think this post does a really good job of covering how multi-pronged performance is: it certainly doesn't hurt uv to be written in Rust, but it benefits immensely from a decade of thoughtful standardization efforts in Python that lifted the ecosystem away from needing `setup.py` on the hot path for most packages.
Yes, sorry! We're investigating, but my current theory is we got overloaded because I relaxed some of our anti-crawler protections a few days ago.
(The reason I did that is that the anti-crawler protections also unfortunately hit some legit users, and we don't want to block legit users. However, it seems that I turned the knobs down too far.)
In this case, though, we had a secondary failure: PagerDuty woke me up at 5:24am, I checked HN and it seemed fine, so I told PagerDuty the problem was resolved. But the problem wasn't resolved - at that point I was just sleeping through it.
I'll add more as we find out more, but it probably won't be till later this afternoon PST.
Edit: later than I expected, but for those still following, the main things I've learned are (1) pkill wasn't able to kill SBCL this time - we have a script that does that when HN stops responding, but it didn't work, so we'll revise the script; and (2) how to get PagerDuty not to let you go back to sleep if your site is actually still down.
Unlike the general public though, these models have advanced dementia when it comes to learning from corrections, even within a single session. They keep regressing and I haven't found a way to stop that yet.
What boggles the mind: we have gone for so long to try to strive for correctness and suddenly being right 70% of the time and wrong the remaining 30% is fine. The parallel with self driving is pretty strong here: solving 70% of the cases is easy, the remaining 30% are hard or maybe even impossible. Statistically speaking these models do better than most humans, most of the time. But they do not do better than all humans, and they can't do it all of the time and when they get it wrong they make such tremendously basic mistakes that you have to wonder how they manage to get things right.
Maybe it's true that with an ever increasing model size and more and more (proprietary, the public sources are exhausted by now so private data is the frontier where model owners can still gain an edge) we will reach a point where the models will be right 98% of the time or more but what would be the killer feature for me is an indication of the confidence level of the output. Because no matter whether junk or pearls it all looks the same and that is more dangerous than having nothing at all.
In a similar spirit there is also a site to scan security headers of any site [1] and another to verify the TLS settings from the Mozilla SSL Configuration Generator [2] and a git repo with code to scan sites from the command line [3] useful if the site is not reachable on the internet or automated scans to HTML reports.
Looking at the htop screenshot, I notice the lack of swap. You may want to enable earlyoom, so your whole server doesn't go down when a service goes bananas. The Linux Kernel OOM killer is often a bit too late to trigger.
You can also enable zram to compress ram, so you can over-provision like the pros'. A lot of long-running software leaks memory that compresses pretty well.
> In other words, immature engineers are too inflexible about their taste. They know what they like, but they mistake that liking for a principled engineering position.
I have come across this although I think quite experienced engineers can suffer from this kind of immaturity.
Aeons ago I used to help friends with their Computer Science assignments. I remember the temptation to rewrite their code because I didn't like the way it worked. I would start to do so and then, eventually, think that it was going to take ages and be unrecognisable to them. How could I help them by rejecting everything they thought?
I'd think about it again and realise that they weren't idiots - their approach probably could work with a couple of adjustments. I helped them make those adjustments and they were happy because they could understand the result.
After that I would find that my own way of thinking about the problem had changed and my own code would get much better from having seen the problem from a different angle. I should really have been the one thanking them.
I am still like this - still prejudiced - but in the back of my mind I know it and when I'm being sensible I remember to try to give the other viewpoint a proper chance and be happy when it really turns out to have more merit.
Principles are a bit subjective and if you lean on them all the time without thought it's a sort of laziness - you're not really examining the situation and what it merits.
- debug@4.4.2 (appears to have been yanked as of 8 Sep 18:09 CEST)
- chalk@5.6.1
- supports-color@10.2.1
- strip-ansi@7.1.1
- ansi-regex@6.2.1
- wrap-ansi@9.0.1
- color-convert@3.1.1
- color-name@2.0.1
- is-arrayish@0.3.3
- slice-ansi@7.1.1
- color@5.0.1
- color-string@2.1.1
- simple-swizzle@0.2.3
- supports-hyperlinks@4.1.1
- has-ansi@6.0.1
- chalk-template@1.1.1
- backslash@0.2.1
It looks and feels a bit like a targeted attack.
Will try to keep this comment updated as long as I can before the edit expires.
---
Chalk has been published over. The others remain compromised (8 Sep 17:50 CEST).
NPM has yet to get back to me. My NPM account is entirely unreachable; forgot password system does not work. I have no recourse right now but to wait.
Email came from support at npmjs dot help.
Looked legitimate at first glance. Not making excuses, just had a long week and a panicky morning and was just trying to knock something off my list of to-dos. Made the mistake of clicking the link instead of going directly to the site like I normally would (since I was mobile).
Just NPM is affected. Updates to be posted to the `/debug-js` link above.
He was a physicist and author with an interest in old technologies. I believe that most of van Leeuwenhoek's microscopes are replicas, but their basic form is well documented. As I recall, the most difficult thing to replicate was the lens, which he made using the same method van Leeuwenhoek's used: dropping molten glass through air.
The telescope was relatively easy in comparison. He used two lens from his stock of lenses and simply mounted them in a tube. I recall him having fun with the decorations, which are quite detailed.
He was so good at making/repairing ancient technology that the Science museum trusted him to repair some of their most precious items from their old wireless collection. He used vintage brass (modern brass looks wrong), vintage ebonite and authenticity old mahogany. Wish I could remember more. I have a few more of his things: a set of Napier's Bones (an old calculation device) made from chop sticks, an reproduction crystal set, a number of orreries.
Firefox simply ignores height declarations that resolve to a value greater than exactly 17895697px. What’s this value? Just a smidgeon under 2³⁰ sixtieths of a pixel, which is Firefox’s layout unit. (It’s the last integer before 2³⁰ sixtieths, which is 17,895,697.06̅ pixels, 4⁄60 more.) I presume Firefox is using a 32-bit signed integer, and reserving another bit for something else, maybe overflow control.
Five years ago, Firefox would ignore any CSS declarations resolving like that, but somewhere along the way it changed so that most things now clamp instead, matching WebKit-heritage behaviour. But height is not acting like that, to my surprise (I though it was).
WebKit-heritage browsers use a 1⁄64 pixel layout unit instead. Viewed in that light, the 2²⁵ − 1 pixels is actually 2³¹ − 1 layout units, a less-surprising number.
You can compute the distance to the moon if you know the radius of the earth by looking at how long lunar eclipses take, data gathered over years of observations.
Eratosthenes computed the radius of the earth by clever trigonometry in ancient times, and Aristarchus computed that a 3.5-hour lunar eclipse indicates that the moon is ~61 earth radii away.
Once you have the distance to the moon, you can compute the size of the moon by measuring how long it takes the moon to rise. It takes about two minutes, and so the radius of the moon is about 0.0002 of the distance to the moon.
By cosmic coincidence, the sun and the moon appear to be approximately the same size in the sky, so the ratio of radius/distance is approximately the same for the sun and the moon. If you measure phases of the moon, you'll find that half moon is not exactly half the time between the full moon and new moon. Half moon occurs not when the moon and the sun make a right angle with the earth, but when the earth and the sun make a right angle with the moon.
You can use trigonometry to measure the difference between the half-time point between new/full moon, and the actual half moon, giving you an angle θ. The distance to the sun is equal to the distance to the moon divided by sin(θ).
To get θ exactly right, you need a very precise clock, which the Greeks didn't have. It turns out to be about half an hour. Aristarchus guessed 6 hours, which was off by an order of magnitude, but showed an important point: that the sun was much larger than the earth, which was the first indication that the earth revolved around the sun. (Aristarchus' peers mostly didn't believe him, not simply out of prejudice, but because the constellations don't seem to distort over the course of a year; they were, as we now know, greatly underestimating the distance to nearby stars.)
Next, you can compute the shape of the orbits of the planets, by observing which constellations the planets fall inside on which dates over the course of centuries. Kepler used this data first to show that the planetary orbits were elliptical, and to show the relative size of each orbit, but with only approximate measures of the distance to the sun (like the θ measurement above) there's not enough precision to compute exact distances between planets.
So, scientists observed the duration of the transit of Venus across the sun from near the north pole and the south pole, relied on their knowledge of the diameter of the earth, and used parallax to compute the distance to Venus, and thereby got an extremely precise measurement of the earth's distance to the sun, the "astronomical unit." It took decades to find the right dates to perform this measurement.
The cosmic distance ladder goes on, measuring the speed of light (without radar) based on our distance to the sun and the orbit of Jupiter's moon Io, using radar to measure astronomical distances based on the speed of light, measuring brightness and color of nearby stars to get their distance, measuring the expected brightness of variable stars in nearby galaxies to get their distance, which provided the data to discover redshift (Hubble's law), measuring the distance to far away galaxies (and thereby showing that the universe is expanding).
We can do better than that! Using the Sun as a gravitation lens[1], and a probe at a focal point of 542 AU, we could get 25km scale surface resolution on a planet 98 ly away. [2] This would be an immense and time-consuming endeavor, but does seem to be within humanity's current technological capabilities.
I often described my wife's old Honda Civic, which we finally sold (still running and able to be driven) w/ just north of 340,000 miles, as having been to the moon and on its way back. I like the idea that someday Honda hardware could, in fact, send something to the moon.
My father was on chemotherapy with fludarabine, a dna base analog. The way it functions is that it is used in DNA replication, but then doesn’t work, and the daughter cells die.
Typically, patients who get this drug experience a lot of adverse effects, including a highly suppressed immune system and risk of serious infections.
I researched whether there was a circadian rhythm in replication of either the cancer cells or the immune cells: lymphocyte and other progenitors, and found papers indicating that the cancer cells replicated continuously, but the progenitor cells replicated primarily during the day.
Based on this, we arranged for him to get the chemotherapy infusion in the evening, which took some doing, and the result was that his immune system was not suppressed in the subsequent rounds of chemo given using that schedule.
His doctor was very impressed, but said that since there was no clinical study, and it was inconvenient to do this, they would not be changing their protocol for other patients.
I use a bunch of marquees to create an animated scene on my homepage[0]. Different speeds for a parallax effect and even some multi-axis marquees for rain effect.
To respond to some of the questions or those parts I personally find interesting:
The custom TUI library is so that I can write a plugin model around a C ABI. Existing TUI frameworks that I found and were popular usually didn't map well to plain C. Others were just too large. The arena allocator exists primarily because building trees in Rust is quite annoying otherwise. It doesn't use bumpalo, because I took quite the liking to "scratch arenas" (https://nullprogram.com/blog/2023/09/27/) and it's really not that difficult to write such an allocator.
Regarding the choice of Rust, I actually wrote the prototype in C, C++, Zig, and Rust! Out of these 4 I personally liked Zig the most, followed by C, Rust, and C++ in that order. Since Zig is not internally supported at Microsoft just yet (chain of trust, etc.), I continued writing it in C, but after a while I became quite annoyed by the lack of features that I came to like about Zig. So, I ported it to Rust over a few days, as it is internally supported and really not all that bad either. The reason I didn't like Rust so much is because of the rather weak allocator support and how difficult building trees was. I also found the lack of cursors for linked lists in stable Rust rather irritating if I'm honest. But I would say that I enjoyed it overall.
We decided against nano, kilo, micro, yori, and others for various reasons. What we wanted was a small binary so we can ship it with all variants of Windows without extra justifications for the added binary size. It also needed to have decent Unicode support. It should've also been one built around VT output as opposed to Console APIs to allow for seamless integration with SSH. Lastly, first class support for Windows was obviously also quite important. I think out of the listed editors, micro was probably the one we wanted to use the most, but... it's just too large. I proposed building our own editor and while it took me roughly twice as long as I had planned, it was still only about 4 months (and a bit for prototyping last year).
As GuinansEyebrows put it, it's definitely quite a bit of "NIH" in the project, but I also spent all of my weekends on it and I think all of Christmas, simply because I had fun working on it. So, why not have fun learning something new, writing most things myself? I definitely learned tons working on this, which I can now use in other projects as well.
Sandwiches, too. Ate at the cafe a fair bit. When my buddy was living on Victory and Hollywood we hit Burbank on a regular basis.
Of the Burbank, Fountain Valley, San Marcos, San Diego, Anaheim Hills, Roseville, Sacramento, Fremont, Las Vegas and Sunnyvale locations, I think I liked the San Diego one most for selection (it was a former Incredible Universe), but the Roseville and Las Vegas ones had the wildest themes, even more than the Burbank UFO. But the Fremont location when I ended up there in 2019 was deader than a doornail, and it was like waiting for the next hit to kill them. The next hit came sooner than I thought.
There's a wonderful book on Mathematical counterexamples in French[0], meant for undergrad/engineering school hopefuls.
So many of the continuity counter-examples are throwing Weierstrass at the wall and getting something to stick. It's fun but also feels a bit like cheating.
I do recommend this book for any french-speaking mathematician-adjacent person though. Real great dictionary for remembering why certain things only work in one direction.
Quick note for anyone who hasn't visited, and has an interest in western civilization: Pompeii, and the somewhat more impressive nearby Herculaneum are well worth visiting at least once. It's really not possible to have the experience remotely from pictures or videos, not the same as being physically immersed. Best to avoid the high summer due to heat and load, but go then if you have no alternative. Herculaneum in particular is never that busy because harder to get to and less publicity.
Yes, definitely. Pi is just the perimeter of the circle, and varpi is the perimeter of the lemniscate. If you use three points, you get three tear-drops, and you can compute the perimeter of that.
Let’s call it a trilemniscate. ;)
Here’s a 3d plot of it. If you rotate to view it from +Z downward, then you’ll see the trilemniscate, which is where the volume intersects with the XY plane. Note I subtracted 1 from the product in order to visualize the plane intersection. (And you can turn off the 3 points version and turn on the 2 points version to compare.)
One interesting note about 2 points vs 3 points. The area inside the lemniscate and trilemniscate is the same! (True for more points, as long as they’re evenly space on a circle). The perimeter, of course, goes to infinity as you add more points.
I'm building Pastmaps - striving to eventually be the world's largest online collection of old maps, aerials, and photos all packaged into a public historical research platform that's as easy to use as Google Maps. This has been a labor of love now for about a year, but I still have a huge mountain to climb to realize the full vision. Give it a try and give me your harsh criticisms - that's the greatest gift you could give me!
Even in it's current state, it's being used by geneologists, urban explorers, search & rescue teams, real estate developers, government agencies, etc. The number of exploding use-cases continues to astound me and keeps me motivated to continue.
I've noticed that my son spends way too much time on YouTube or playing Minecraft and one of the few offline activities he enjoys doing on his own is coloring. And since he comes to me every time he wants a new coloring book and we spend about 10 minutes together searching for each picture, I made a website with a collection of coloring books for him. The site is very simple, but to be honest, I haven't had so much fun with the process of creation for a long time.
I can do more with the Internet today than I could with a static /22 assigned over my ISDN BRI back in the mid-1990s. A lot of things I would do back then, I would do differently today; running a chat system by connecting directly out to 6667/tcp feels pretty silly now, for instance. It's rough to build protocols that work that way today, but you're not missing much. Things were not better before the advent of presumptive NAT.
We’re dealing with this big time in Asheville now. When cell service came back at all, everyone had shitty intermittent 3G, and none of the websites we needed for basic survival information would load. A bunch of good people created some text only news sites, and today I noticed that the Buncombe county website finally has a low bandwidth site, but even then when I inspected it, it had 130k of bootstrap css and 50k of jQuery blocking rendering. It’s great that people are doing this work, but citizens needed this a week and a half ago. By now, I’ve figured out where to get water, food, non potable water, etc. Seeing tech fail so badly through all this has been eye opening for me, in a depressing way.
Ron Miller is an artist who made some very nice visualizations. I can’t vouch for the scientific accuracy, but they seem plausible enough to me, and consistent with the images I’ve seen of Saturn’s rings from nearby probes.
"At the current rate of clearance it is a conservative estimate that the Département du Déminage will still be finding these weapons nine hundred years from now."
I was the person who first deployed Python at Goldman Sachs. At the time it was an "unapproved technology" but the partner in charge of my division called me and said (This is literally word for word because partners didn't call me every day so I remember)
Err....hi Sean, It's Armen. Uhh.... So I heard you like python.... well if someone was to uhh.... install python on the train... they probably wouldn't be fired. Ok bye.
"Installing python on the train" meant pushing it out to all the computers globally in the securities division that did risk and pricing. Within 30mins every computer in goldman sachs's securities division had python and I was the guy responsible for keeping the canonical python distribution up to date with the right set of modules etc on Linux, Solaris and Windows.
Because it was unapproved technology I had a stream of people from technology coming to my desk to say I shouldn't have done it. I redirected them to Armen (who was a very important dude that everyone was frightened of).
The core engineers from GS went on to build the Athena system at JP and the Quartz platform at BAML.
ChatGPT is a useful product, which they're monetising in a well-travelled internet company way. The bad news is you're going to have ads in your ChatGPT in 2030. The good news is you're still going to have a job in 2030.