w-m

Another curious data point: id =40000000 is up in a few hours. I thought it would be fun to try to make a submission pointing to that id, while trying to get that id right when it’s up. But it’s too late to start hacking on a script; I’m rather going to bed now. (Anybody else trying, please don’t kill the server or make dang mad).

Edit: oh I missed a 9 in looking at an earlier link, it’ll be a little more than a day from now, unless sama announces a new product, triggering a comment storm.

ffmpeg updates their API very liberally, even in minor version bumps. Combine that with many packages depending on ffmpeg doesn't make it very easy to always have the latest ffmpeg version. They are working on it though: https://trac.macports.org/ticket/65623

(I'm not a MacPorts maintainer, but I've been burnt by ffmpeg API changes a couple times myself before).

Is there any list that's growing slower over time than the Dedekind numbers? https://oeis.org/A000372

New entries in the years 1897, 1940, 1946, 1965, 1991, 2023.

(see also "Ninth Dedekind number discovered: long-known problem in mathematics solved" https://news.ycombinator.com/item?id=36491677)

The original 3DGS paper offers a download of a few scenes (14 GB): https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/dat...

The link is taken from the projects' GitHub page: https://github.com/graphdeco-inria/gaussian-splatting

For some reason the GUI is not showing up for me in 3DGS.cpp. (I checked out the repo, made a build folder and built with Ninja, then launched ./apps/viewer/vulkan_splatting_viewer).

I still have a VulkanSplatting build from this Wednesday though. In VulkanSplatting, when looking at the Lego scene from above with the default window size, I'm getting just below 1 ms for the sorting kernel and just above 1 ms for "render", everything else is too small in the graph to register. But it only displays at a handful of fps, so it seems quite some time goes unaccounted for.

Maybe spinning up Instruments could give some more insights into what's happening? I tried `cmake -G Xcode` to have that setup easily, but the Xcode CMake generation fails with

    CMake Error in src/shaders/CMakeLists.txt:
      The custom command generating
    
        3DGS.cpp/build-xcode/shaders/shaders.h
    
      is attached to multiple targets:
    
        shaders
        xcode_shaders
    
      but none of these is a common dependency of the other(s).  This is not
      allowed by the Xcode "new build system".

Great project, it's nice to see so many people are building stuff with Gaussian Splatting.

Just the other day I went through the whole list of known viewers on MrNeRF's awesome 3DGS resources, to find one that runs on a MacBook. I'm working on compressing 3D scenes by sorting Gaussians into 2D grids [0], and I wanted a native viewer that I could for experiments on the go.. perhaps as an alternative backend to the CUDA one in my colleague's exploratory Python viewer [1].

VulkanSplatting was the only one I could get to compile and run on my Intel MacBook. Unfortunately the feeble Intel GPU isn't able to display even the Lego scene at an interactive framerate. Do you think there's performance headroom, and that it will become possible in the future, or should I give up trying to run this on an Intel MBP?

[0]: https://fraunhoferhhi.github.io/Self-Organizing-Gaussians/

[1]: https://github.com/Florian-Barthel/gaussian_viewer

I was thoroughly confused reading the Dada Manifesto[0], starting with the non-existent German meanings of the word dada and getting much stranger from there. Until I found out at the very bottom that it's a riff on a 1916 dada manifesto.

[0]: https://dada-lang.org/blog/manifesto

I've had that same message a couple of times over the last few days, while logged in (EU).

It's kind of expected behavior when there's a thread with thousands of comments (like when OpenAI shafts their CEO [0]), but the last few days seemed relatively calm.

So maybe the server is crumbling under the increase in traffic from higher HN rankings in Google [1]?

[0]: https://news.ycombinator.com/item?id=38310213

[1]: https://news.ycombinator.com/item?id=39423949

> how do you use -0 in the add

In my mind the two zero values would represent a tiny epsilon around 0, let's say -0.01 and +0.01. Looking at them like this, it would mean

  +0 +0 -0 = +0
  +0 -0 -0 = -0
  +1 * +0 = +0
  -1 * +0 = -0

Performing addition with the same sign count in each group would be problematic. How to decide on the sign of +0-0 or +1-1, other than flipping a coin?

I was reading Exposing Floating Point today (as Airfoil is on the HN front page and I was perusing the archive of the author). It's a blog explaining the inner workings of floating point representations. About zero values it says [0]:

> Yes, the floating point standard specifies both +0.0 and −0.0. This concept is actually useful because it tells us from which “direction” the 0 was approached as a result of storing value too small to be represented in a float. For instance -10e-30f / 10e30f won’t fit in a float, however, it will produce the value of -0.0.

The authors of the LLM paper use the values {-1, 0, -1}. Connecting the two ideas, I'm now wondering whether having a 2-bit {-1, -0, 0, 1} representation might have any benefit over the proposed 1.58 bits. Could the additional -0 carry some pseudo-gradient information, ("the 0 leaning towards the negative side")?

Also, I've seen 2-bit quantizations being proposed in other LLM quantization papers. What values are they using?

[0] https://ciechanow.ski/exposing-floating-point/#zero

And at the same time borrowing the name and file ending of Pythons object serialization protocol.

Sales weighted battery capacity has nearly doubled since 2021: https://www.adamasintel.com/average-battery-capacity-converg...

The Tesla Model S was introduced with 60 and 85 kWh batteries ~10 years ago. Today you can buy vehicles in the same class with 120 kWh, and dozens of models with 85+ kWh: https://ev-database.org/cheatsheet/useable-battery-capacity-...

Of course affordability remains a problem, but the battery capacity per vehicle has drastically improved, even without world shattering breakthroughs in chemistry.

From the article:

> There was no such thing as an average pilot. If you’ve designed a cockpit to fit the average pilot, you’ve actually designed it to fit no one.

There was a comment today talking about this, and I hadn’t heard the story, so I looked it up.

https://news.ycombinator.com/item?id=39110365

If your language has its own word for the concept of taking pleasure in other people's misfortunes, you don't need to have any great comedians (left [0]) to be able to laugh.

Part of the family of haha, hehe and hihi is höhö in German, which can be used as a sarcastic laugh to a particularly bad joke, halfway between a ho-ho-ho and a hehe.

Also when speaking out loud, the pronunciation of lol in German is always "lol", as a word. I believe in English-speaking countries the letter by letter l-o-l is more common.

[0]: https://www.youtube.com/watch?v=qO1No7HWT6Q

That would be much appreciated. I've sent a message to support about the email not being filled in by Keychain on iOS, request 4054436.

Messaging support in this way would make sense to me for reporting a weird, rare bug. But what I described is widely seen in the community[0-3].

In fact, looking for these links, I now found an official Notion reply from 4 years ago in one of these threads[4]: "should keep you logged in for 90 days per device".

This had me find [5]: "Notion has a default session duration of 90 days. This means all users automatically get logged out if they have stayed logged in for 90 consecutive days.".

If logging out any session after 90 days is the expected behavior, this would completely explain the behavior in my case. Can you confirm that this is happening for all accounts?

[0]: https://www.reddit.com/r/Notion/comments/brcy5m/random_logou...

[1]: https://www.reddit.com/r/Notion/comments/vi8c9f/constantly_g...

[2]: https://www.reddit.com/r/Notion/comments/194xw0m/anyone_else...

[3]: https://www.reddit.com/r/Notion/comments/dsi5jo/login_proces...

[4]: https://www.reddit.com/r/Notion/comments/dsi5jo/login_proces...

[5]: https://www.notion.so/help/hipaa

In case anybody from Notion is reading here: I'm very frustrated by being logged out on all my Apple devices (MacBook, iPhone, iPad) every few weeks to months.

By itself this would be only annoying, but on top of the spurious logout, none of the Notion apps support Keychain access. Thus when having to log in (with my email and password combination), I need to hunt for those in the password list and copy/paste them, which takes quite a bit of time doing that on each device.

Google searches tell me that I'm far from alone in experiencing it and that the problem seems to exist for several years.

Tangent on your tangent: I always find it awkward that people name the Wright brothers as the inventors of the airplane, and keep forgetting Otto Lilienthal.

The Wright brothers engineered a control system, that’s nice and all, but pretty boring, iterative work. Lilienthal studied birds, wrote the book on how wings work, produced the first airplanes and was the first person to make well-documented, repeated, successful flights with gliders. In my view these were much more difficult steps and much more important work.

Lilienthals life is super interesting, but even his namesake museum in his hometown of Anklam is rather pitiful.

I’ve seen “anywhere in earth” [0] used increasingly for deadlines. It’s something I’m looking up in these time zone calculators. You might consider adding it.

[0]: https://en.wikipedia.org/wiki/Anywhere_on_Earth

I would expect something like a new VW car interface, that is put into millions of cars, to pass some kind of internal design review. How do these things (like touch controlled AC sliders without backlight) get introduced, which would take two minutes of usability testing to be thrown out? Can this just be explained by cost savings, or is there something else going on?

I went in with an expectation of it being completely different. And there is much more manual labor in 1939. It's dirtier, and grittier, but mostly for the steps we don't see in the Toyota video, as for example the motor comes in one block, and we don't see how it's built.

But on a high level, both videos look like the same process. You can find many steps, like the stamping, the welding, the final assembly, that look nearly unchanged in 85 years, save for some automation with robots, and some quality-of-life improvements in the assembly, where people don't have to lug around heavy parts.

From yesterday: AMD MI300 performance – Faster than H100, but how much?

https://news.ycombinator.com/item?id=38550271

Picked one from my recent history: https://www.videogist.co/videos/10-things-you-didn-t-know-ab...

The text summary is very, very good. Kudos!

(As a nitpick, it has 'Wesan' instead of 'Wunderschön', but David himself doesn't pronounce it very clearly, so can't really blame the speech recognition there.)

The frame grabs next to the text appear to be sampled at random times, and not for their content. In the video, David presents screenshots and photos and what not. It happens to work out for Chapter 4, 6 and 7. In 5 of the other chapters it shows just the talking head. If a human had picked them, I think they'd show the most relevant imagery for the particular part of the video, and not the talking head.

I’m not sure who “they” are, but I’m pretty sure they’re already doing that, and don’t need me to get the idea. I think it’s important to talk about what LLMs mean for privacy. Profiling every HN user might be a useful tool to are people more aware of the problems. But I totally get your unease, which is also why I haven’t done that myself.

The cat is out of the bag, can’t get it back in by ignoring the fact.

Sure, everything is ultimately a political problem, but this one is completely driven by technological change. In the USSR (and GDR), it took them a staff of hundreds of thousands of people to write up their reports.

Now it would take a single skilled person the better part of an afternoon to, for example, download a HN dump, and have an LLM create reports on the users. You could put in things like political affiliation, laws broken, countries travelled recently, net worth range, education and work history, professional contacts, ...

Waiting for amber takes 20 seconds. Also it’s not a timeout of some case from repeated use. Have to wait even if it wasn’t used for several minutes beforehand.

Seems a good place to ask a question I’ve been asking myself for a long time, and couldn’t figure out the answer to: why do lonely crosswalks not immediately turn the lights red for the cars when you push the button? Or at least, after a very short delay (< 3 seconds).

For street crossings with multiple directions and multiple lights, that behavior makes sense. You’re waiting for the next pre-programmed phase that syncs you up with other parts of the crossing. For large roads it might make sense, you don’t want to interrupt the flow of the traffic that is expected to hit a green light down the road at current speed.

But for the lonely crossing light in front of the school building in a small street, why do I have to wait 20 seconds after pushing it, every single time?

You can create named API keys, and easily delete them. Unfortunately you can't seem to put spend limits on specific API keys.

If you're not using the API for serious stuff though it's not a big problem, as they moved to pre-paid billing recently. Mine was sitting on $0, so I just put in a few bucks to use with this site.

Playing with opposites is kind of fun, too.

Simply a cat, evolving into a lounging cucumber, and finally opposite world:

https://dalle.party/?party=pqwKQVka

Vibrant gathering of celestial octopus entities:

https://dalle.party/?party=lHNDUvtp

Another data point, the top AskReddit posts of the past year [0].

Zero of the top 25 posts of the last twelve months are younger than five months. Two of the top 50 posts are younger than five months.

Comments (using subredditstats from a sister comment) have gone from around 70k per day to around 12k per day.

[0] https://old.reddit.com/r/AskReddit/top/?sort=top&t=year