This whole blog is fascinating, assuming that you're interested in the intersection between (recent) history, demographics, politics, and religion in the US.

For example - this one about generations and trust was surprising to me: https://www.graphsaboutreligion.com/p/gen-z-doesnt-trust-any...

or this on perception of the religiosity of holidays: https://www.graphsaboutreligion.com/p/is-christmas-a-religio...

Or only use the subscriptions page, and never visit the home page.

Getting only the stuff you've curated is way better than the algorithm.

I'm one of those that instinctively types the letter "y", presses the down arrow, then enter, and goes straight to the subscriptions page.

Unfortunately, I now have to scroll through the "most relevant" section, then through shorts, and only then do I reach the videos in the chronological order. What I also hate is how when a channel I usually engage with uploads a video I have no interest in, then it's stuck on top of my subscription page in that "most relevant" section for like a week.

You can use Firefox plugin "Unhook remove youtube recommended and shorts" I think I also have some ublock filters that blocks those things. So, I only have videos.

you can reduce shorts for a period of 3 weeks or so,

When I was building a linux PC in the late 90's, I bought a Matrox Millenium with the memory expansion to 8MB which could do 1600x1200 with 24 bit color.

That was an awesome card for the time, and an incredibly stable analog signal, which Matrox was known for - other cards at the time would get fuzzy, shimmer, or have ringing / ghosting.

At the time, for good support you needed to *buy* an X11 server for accelerated graphics support. I forget which one I paid money for, but within a few years it wasn't necessary as open source had caught up.

Yeah, I wanted a Matrox so bad at that time because you could also do dual monitor on some of those cards. I had a sick NEC with a degauss button and trinitron style horizontal lines and a shit packard bell monitor. I wanted to do dual monitors so bad with both, despite the resolution asymmetry. Some matrox also had connection points for antennas and video capture. They were truly workhorse cards. But they sucked at 3D and were lapped in the Boot mag benchmarks for Quake 3D et al. I ended up with a 3D Labs Permidia 3D card anyway because I couldn't afford a Diamond or Voodoo and ended up turning textures off fully on Quake 3d in the end. Fun times, shout out Hard OCP.

I had dual Trinitrons around that era, to this day when I build myself a new computer desk I massively over build it (the latest one I can stand in the middle without noticeable flex..) because I lived in slight dread that the spacetime warping monitors would crush the one I had at the time.

I remember people I knew wanting the Matrox because it was possible to have duel screen on Linux. They usually had a huge grin on LAN parties, once they got their breath back from carrying two heavy CRT monitors and a PC tower. Good times.

Maybe I misunderstood something at the time, but the mga X11 driver was weird. My understanding, at least at the time, is that it is/was possible to provide an X11 driver that would work regardless of the underlying operating system, providing features such as dual-head support.

This feels intuitively wrong to me now and I doubt that's how it worked. In any case I dutifully copies the Matrox provided X11 driver to OpenBSD and enjoyed multi-monitor support.

There was nothing underlying X servers until about 2010. The X server was the lowest layer.

Looking at a few old mailinglists agrees with you. Apparently the mga_drv and mga_hal_drv could just be copied to any system running XFree86, at least Linux and the BSDs, and would let you utilize dual-head support.

I think I had a Mystique which may have been a PCI model as well. I remember the signal quality being very visible switching a display from that box into some other one which had some S3 or something. The image looked noticeably blurry or fuzzy with the exact same display and cables on the worse card.

You mean the one from Xi graphics?

That's Accelerated/X right. I had a Matrox Millennium card in my PC when I discovered Linux and remember trying that. Made a lot of difference.

Dupe of link to original post on reddit: https://news.ycombinator.com/item?id=49216926

My first desktop linux box was a PowerMac 8100/80 with MkLinux, probably sometime in 1996. At launch it didn't have support for multi-button mice, which made X11 kind of hard to use (center-button was the only way to paste text in many cases back then), as Apple only shipped a one button mouse by default so that was all that was supported.

As an impertinent high schooler, I emailed Logitech asking for one of their 3 button ADB trackballs, saying I was going to add 3 button support to MkLinux. They actually replied and sent me one. I managed to come up with some code that kind of worked, and sent off a patch, but by that time I think someone else fixed the multibutton mouse issue before me.

Good times back then. It was such a step up in terms of multitasking, and a step back in terms of UI consistency from classic MacOS.

When I went off to university a year later, Apple wasn't doing so hot so, so I built myself a Pentium Pro machine. Prices on those parts went through the floor as the Pentium II had come out, and Linux was all 32 bit so no 16 bit perf issues the PPro was known for. Dual CPU board (single CPU at first) and as another processor and memory upgrades getting cheaper it was a nice machine through all of college.

Then after graduation I got a job after MacOS X came out, and jumped back to Apple products as they were the only game for unix on a laptop with power management that didn't suck in the early 2000's.

Good times.

Good times indeed. Similar to you, my first Linux install was MkLinux on a PowerMac 7100. And I had completely forgotten about the mouse button issue until you mentioned it. I vaguely recall somehow attaching a 3-button PC serial mouse and writing some code to make all the buttons work.

Power management still sucks on Linux AND windows for that matter. Ridiculous how every other UX is just left to rot, it’s like no one else even tries besides Apple.

I'd take this the opposite direction - HTTP should get more like SMTP.

Take this thought experiment - how does nearly every other standard protocol (DNS, SMTP, etc.) except for HTTP survive at scale without server-side load balancing?

Thankfully we're heading in this direction due to happy eyeballs and related client-side retry mechanisms.

Have you benched this for coding tasks, with a fallback to a larger local model, for example Qwen-3.6-27B?

Or using it for sub-tasks, where a framework with a larger primary model dispatches simpler jobs ("summarize ...", etc.) to it?

We are currently working on larger models atm! your suggestion is cool btw

A YouTuber automated ROCm version testing for Strix Halo/R9700 + LLM workloads:

https://kyuz0.github.io/amd-strix-halo-toolboxes/

https://kyuz0.github.io/amd-r9700-ai-toolboxes/

What hardware (CPU/GPU/memory) and network was used for this? What quantization for GLM 5.2? How much tuning of the split was needed?

The lab features two Mac Studios: an Apple M3 Ultra (32 CPU cores, 80 GPU cores, 256 GB unified memory) and an Apple M1 Ultra (20 CPU cores, 48 GPU cores, 128 GB unified memory), both connected via 1Gbit Ethernet.

We use a customized Q2 quantization that preserves sensitive tensors at Q8.

To reduce compute time per layer, we are developing a custom GLM DSA Metal graph.

While we are not yet approaching MTP, we plan to port our existing MTP implementations from versions 4.7 and 5.1 to 5.2.

Since GLM's MTP acceptance rate is very high for a single predicted token, we are exploring token prediction techniques to widen the predicted tokens and utilize parallelism for verification.

Equivalent M3 machines no longer for sale from Apple (only up to 96 GB) but can be had on eBay for around $14,000 each

It's notable that they're so valuable because they feature 800Gbps of memory bandwidth. About twice what's available on the top end of M5, and exactly what makes llm inference fast.

> because they feature 800Gbps of memory bandwidth. About twice what's available on the top end of M5

Ouch, about half of the memory bandwidth of a dedicated GPU though :/ Running LLMs on Apple hardware still doesn't make any sense to me.

M5 max has 614GB/s, you mean the m4?

Oh, I was looking at every M5 except for the 40-core M5 Max. They have 460.

While Thiel's recent Antichrist talks could come to mind, this could just as easily be business speak, like in the Collins/Porras "Built to Last" where one of the points of great companies is "Cult-like culture".

Yes. There are companies which want "passion", which is really about employee motivation. Zappos, which is a shoe store, has been laughed at for this.