datenwolf

@freakboy3742@cloudisland.nz If you're putting lots of text on your slides, you're doing it wrong anyway. The slides are stage dressing for your talk, they NOT ARE your talk! Put in your slides which can't be communicated verbally. Like photos of your experiments, data graphs, figures, the likes. Example:

@integerpoet@sfba.social @drahardja@sfba.social Not so much differentiate it, but more like make it sellable at all: Smartphone cameras are optical garbage; if you'd look at the output of a smartphone camera that was processed in a traditional way, you'd barf all over it. They're terrible. Only by sending them through the indigestion of an image enhancement transformer network they become palatable. The downshot of that is, that consumers who don't give a fuck just accept it, killing the market for proper cameras.

@ErikUden@mastodon.de Great… another item on the list »Finally, we've built the Torment Nexus, conceived in the novel "Do not build the Torment Nexus!"« checked. ☑️ Real life implementation of the "Brains in Vats" horror subgenre EDIT: If you're in need for a chuckle, here you go: https://qntm.org/dropout

@vampiress@eigenmagic.net Yes, and also: Every line of code you consciously decide not to write is an active design decision (in the same way deleting code is) you ARE making.

@futurebird@sauropods.win Ha… these days it's become practically impossible to develop a new internet search from scratch: You've to crawl the web to build an index. Which means that you've got to scrape the web just like the AI crawlers do. Which means that these crawlers will run into the same self-defense mechanisms that were rolled out to have at least some protection.

@jiub@not.an.evilcyberhacker.net or switch to https://vim-classic.org

@theorangetheme@en.osm.town @davidgerard@circumstances.run already happening. A friend of mine is already selling a certain kind of shovels for the whole upcoming quantum hype.

@cczona@hachyderm.io @pojntfx@mastodon.social I wonder how long it will take for the rest of the world to a) realize what's going on, b) arrive at the objectively valid conclusion and c) act on it and see it through. Recently past experience (COVID-19) shows that being a tall order. Anyway, I wonder when, if so, people who actively rejected contact with LLMs will be in extremely high demand, being knowledge bearers in teaching and technical lead positions.

@anthropy@mastodon.derg.nz @Em0nM4stodon@infosec.exchange The dotcom bubble at least have us some lasting infrastructure. Much of the fiber the "regular" ISPs (not MAANG) operate on are legacy from the hyped up build-out of the dotcom bubble. The AI hyperscaler data centers OTOH: Their "G"PUs are worthless for anything else other than super-low precision sparse matrix multiplication. Can't even use them for low stakes DSP. The networking stuff might have some aftermarket use. But the rest? e-waste.

@Dtraslerwriting@mastodon.social @niekvleeuwen@mastodon.social @dbattistella@mstdn.ca I'm reasonably sure individuals with those psychological traits did exist back then, too. They just had an abysmally low long term life expectancy, due to getting kicked out and exiled by the rest of the commune.

@brouhaha@mastodon.social @purple@nya.social @futurebird@sauropods.win @fivetonsflax@tilde.zone @oscherler@tooting.ch @argv_minus_one@mastodon.sdf.org @CursedSilicon@social.treehouse.systems As a laser physicist specialized in high speed tunable fiber lasers, let me put it this way: Don't mix water and optical fiber. It does funny things. https://en.wikipedia.org/wiki/Hydroxyl_ion_absorption

@mia@movsw.0x0.st @navi@social.vlhl.dev @neal@social.gompa.me @dalias@hachyderm.io @miah@hachyderm.io X.org is not a special citizen. For a long time it's been using the very same interfaces to kernel land as Wayland-land does as well (DRM). If X.org can lock up drivers in kernel code, so can Wayland. Which is then a bug that should be fixed DRM side in the kernel.

I'm getting annoyed with the asymmetry of the PC104/105 keyboard arrangement. I guess the time has finally arrived for me to design and build my own, custom keyboard.

Wishlist:

• movement block ( Ins, Del, Home, End, PgUp, PgDown, ←,↑,↓,→ ) and media control keys on the left

• Letters/symbols/digits and F-keys goes into the middle

• Split main block for angled ergonomics?

• TrackPoint?

• Numeric entry block on the right.

Also, I do enjoy writing gateware more than writing firmware → FPGA.

@purple@nya.social @brouhaha@mastodon.social @futurebird@sauropods.win @fivetonsflax@tilde.zone @oscherler@tooting.ch @argv_minus_one@mastodon.sdf.org @CursedSilicon@social.treehouse.systems The ICs used for the frontend are the same on either end. The difference is, that in a DSLAM a single subscriber card usually serves between 8 to 16 customers, and internally has a couple of relays for every line to swap to reserve frontends. You can buy them in two "regular" and "extended" temperature range, the later being used in DSLAMs, because those might experience intense sunlight and raised temps.

@anthropy@mastodon.derg.nz @Em0nM4stodon@infosec.exchange Only recent versions of the tensor cores do support the low precision formats. OG tensor cores (and TPUs) for that matter, would have best performance with fp16 but also support "pseudo" fp32 (same range but less precision) and even fp64 albeit with much reduced performance. These are the useful number formats for scientific and simulation stuff like weather and climate modeling. The low precision formats are a rather new addition and pretty useless for DSP & sim.

@s4mmy@chaos.social Oder man geht direkt zur Quelle und zieht sich die Daten von den Servern des US-amerikanischen Nationalarchivs: https://catalog.archives.gov/id/12044361 Ich habe da gestern einige Ur-(Ur-)-Verwandtschaft gefunden; da ich gerade bei meiner Mutter (die momentan an einer Familienchronik arbeitet) zu Besuch bin gezeigt und sie so: Ah, von "dem" habe ich die Originaldokumente – holt einen Ordner aus dem Regal, klappt auf und hervor kommt eine Urkunde mit dem Nazi-Reichsadler oben drauf.

@notjustbikes@social.notjustbikes.com @uxMark@cosocial.ca I wish we here in Germany could adopt this mentality. But as it is, even the most well meaning attempts at "improving" bicycle infra are haphazard, any may even be worse, than was before. Take my home city of Lübeck for example. They're finally (with a 5 year delay) building the new cycle-expressway. Unfortunately they completely misunderstood the "continuous sidewalk" portion of the assignment. "car-brain pacifying" adjustments were made, which IMHO endanger cyclists.

@ricci@discuss.systems @freakboy3742@cloudisland.nz It's not visible from these static screenshots, but: My slides never reveal all information at once. I usually create a logical choreography, where one piece of information is revealed at a time. For example in the first slide I started with a picture of the unmodified microscope head, explaining the engineering problem (need to get our scanning beam in there). Then reveal the optics, followed by the measurement showing how well it integrates. And at last, the real thing.

@SeaDonut@tech.lgbt @Kroc@oldbytes.space Modern GPUs are power hogs. In terms of pure 2D graphics that doesn't apply convolving effects, a well optimized CPU implementation can easily outperform a GPU in terms of power consumption and performance.

@mia@movsw.0x0.st @navi@social.vlhl.dev @neal@social.gompa.me @dalias@hachyderm.io @miah@hachyderm.io Not using the GPU for simple stuff is something I've learnt when I was looking to add vector graphics to my scientific visualization library, and it turned out that just using Blend2D to render the vectors on CPU, then upload to the GPU and composite with the GPU rendered stuff was more efficient (quicker and used less energy) than trying to coax the GPU into doing it.

@brouhaha@mastodon.social @purple@nya.social @futurebird@sauropods.win @fivetonsflax@tilde.zone @oscherler@tooting.ch @argv_minus_one@mastodon.sdf.org @CursedSilicon@social.treehouse.systems In fall 2025 / spring 2026 a lot of hollow channel conduit was installed under the sidewalks in the streets neighboring the house I live in. So I got a very good look at how those lowest-bidder-contractors handled it. I was disgusted and appalled. And that was before they even got to running the fiber.

@brouhaha@mastodon.social @purple@nya.social @futurebird@sauropods.win @fivetonsflax@tilde.zone @oscherler@tooting.ch @argv_minus_one@mastodon.sdf.org @CursedSilicon@social.treehouse.systems A lot of the engineering that goes into manufacturing optical fiber cables for outdoor and/or buried installation goes/went into the water barriers. Yes, it should be fine for the coming decades. Just don't expect the same for 900µm buffer fiber pulled through burried hollow channel conduits. I mean the conduits by themselves are fine, but the way they're installed…

@purple@nya.social @brouhaha@mastodon.social @futurebird@sauropods.win @fivetonsflax@tilde.zone @oscherler@tooting.ch @argv_minus_one@mastodon.sdf.org @CursedSilicon@social.treehouse.systems However, in my personal experience it's not raised temperture that's degrading my DSL connection, but rain. I suspect the copper running from the house I live in to the closest DSLAM is over 60 years old and the isolation already brittle and cracked up in places, so that it sucks water, so stronger losses and other fun degraded transmission line effects.

@whitequark@social.treehouse.systems I spent more than a few brain cycles on thinking about, how far you could shrink thyristor-like structures in order to implement high density, low structure count SRAM.

@azonenberg@ioc.exchange @da_667@infosec.exchange You can do that. Google "cargo cruises" or "cargo voyage"

@davidaugust@mastodon.online @SecondUniverse@autistics.life Nah… I say we implement the mythical sacrifice of "the most virtuous (virgin)" to appease "the gods". Every month we kick the most virtuous = wealthy down into an active volcano, in order to appease the god of the economy, the "market". I'm willing to bet that doing so actually had a measurably positive net effect on the economy and societies.

@HydrePrever@mathstodon.xyz @freakboy3742@cloudisland.nz My talks usually contain several "catch up" slides where the information of the previous 5 slides give or take is summarized. I'm doing these also for personal convenience, since often I'm asked to give "elevator pitch" presentations on the same topic, and then I can just use those summary slides instead. They're not "lead wastelands", too, though. Mostly visual communication, zero prose, Just key data and bullet points.

@wonka@chaos.social @masek@infosec.exchange @EricAlper@mastodon.social Add me as well.

@puppygirlhornypost2@transfem.social Every once in a while I vocalize my mental state (more often than not a state of frustration and confusion) in a single monosyllabic grunt. When asked: "What's up?" my usual response is "Do you *really* want to know?" will be acquitted by "I don't have the mental capacity for one of your in-situ lectures/infodumps"

@ohir@social.vivaldi.net @killick@dmv.community @freakboy3742@cloudisland.nz Not ever did I get any money for giving talks at conferences. As a matter of fact, one usually has to pay a substantial registration fee (typically somewhere in the range 300USD to 900USD, depending on conference).

@the_wub@mastodon.social @tokyo_0@mas.to Back in the 486DX2 / Pentium-1 days the main limit was available memory; swap space a band aid just so you could actually run software. Back in those days you actually had to terminate programs just to free up resources if you wanted to do something else. Even today I maintain the opinion, that programs should be developed with a mindset, that starting them should be nearly instantanuos. 1/

@dat@social.g33ky.de Well, okay then.

@cadey@pony.social "Ah shit, here we go again…" #linux #lpe #0day

@Netzblockierer@tech.lgbt @erd@gei.st Stimmt alles. FlatPak / AppImage "lösen" ein anderes Problem, spachteln dabei aber über das generelle Problem drüber, sodass es nicht auffällt. Der Hauptproblem ist, dass unter Linux und den *BSDs aus unerfindlichen Gründen die C-Standardbibliothek (libc) die Aufgabe des Syscall-Trampolins mit übernimmt. Das hat schon mehrfach zu Problemen geführt wenn eigentlich völlig unrelatierte Dinge (z.B. statisch pthread Mutex initializer) nicht mehr passen (siehe glibc-2.41)

@mia@movsw.0x0.st @navi@social.vlhl.dev @neal@social.gompa.me @dalias@hachyderm.io @miah@hachyderm.io Yeah, I already mentioned Slug. And for some time I was cooking up something similar with my Rund glyph rasterizer (maybe I'll finish that as well, just out of spite regarding how EL published Slug). But it's missing the elephant in the room: Using the GPU is wastefull. Each and every process that loads GPU APIs will end up with >100 MiB of non-deduplicable (b/c ASLR) redundance in their resident working set. https://dl.datenwolf.net/vkdlopen.c

@Netzblockierer@tech.lgbt @erd@gei.st

Der WinNT-Kernel (also nur der Kernel, und nicht die darauf aufbauenden "personalities") ist von der Architektur und der Implementation her ziemlich nice (waren ja auch extrem erfahrene und fähige Leute die das Ding entworfen haben).

Win32 als API ist auch… akzeptabel.*

Aber der ganze Rotz der darauf aufbaut? Ja, kann weg.

–

• es wird behauptet (und kann wenig dagegen sagen), dass Win32/WINE/Proton die beste Anwendungs-API für Linux sei.

@dat@social.g33ky.de If I'm doing this, it'll become 4 PCBs of similar dimensioned pairs, instead of a single one with high aspect ratio, just for easier (=cheaper) manufacturing. Can then arrange the blocks in whatever way in a 3D printed chassis.

@mia@movsw.0x0.st @navi@social.vlhl.dev @neal@social.gompa.me @dalias@hachyderm.io @miah@hachyderm.io I do believe you. I know how the DDXs do work. And it doesn't matter that it's X.org stalling the GPU. Anything that can open /dev/dri/card* can. And as I said: Using GPUs for simple graphics stuff (and everything you can do with X11 core and Xrender is simple) is wasteful. I keep repeating myself: Do not use the GPU unless you absolutely have to. And that extends to the X11 server.

Uh… #bandcampFriday quality-of-life question: Is there some convenient tool (preferrably CLI), that will automatically download everything one has on account? Ideally it would also look in one's local collection and only download what's not there, yet.

@Erebus_Amauro@ohai.social @ilumium@eupolicy.social @EUCommission@ec.social-network.europa.eu Same. Just to occupy the namespace I'll register an account there, create a pinned post explaining why I'm NOT there with a link away to my actual profiles. Then throw away the key.

@Netzblockierer@tech.lgbt @erd@gei.st Es wäre schon sehr viel gewonnen, wenn die ganzen Syscall-Trampolins/Wrapper nicht Teil der libc wären, sondern z.B. durch den Kernel selbst als VDSO in das Prozessabbild linked würden. Die libc ist (steckt ja im Namen) die Standard-C-Bibliothek. Braucht man also gar nicht für Binaries die nicht mit C oder C++ kompiliert wurden (ausser die ziehen sich das über eine Abhängigkeit rein).

@alwayscurious@infosec.exchange @whitequark@social.treehouse.systems Yes, current equals power dissipation. You want something that keeps a charge. Which is exactly what SRAM does. But it requires a lot of structures to do so.

@ricci@discuss.systems @freakboy3742@cloudisland.nz Think about this: Take a video of 3blue1brown for example. The visuals are the "slides", and the verbal explanations are the talk. Is the video taking attention away from the text spoken by Grant? Or vice versa? No, they work hand in hand.

@the_wub@mastodon.social @tokyo_0@mas.to

When it comes to UI design "suprising" behavior – which includes suprises as "oh, you can do THAT?" – should be avoided.

1. it creates a rarely trodden usage path, which means user interaction bugs may remain undiscovered for a long time.

2. it creates support pressure, if you have to explain umpenth times to users how to reach certain information.

2/

@anthropy@mastodon.derg.nz @Em0nM4stodon@infosec.exchange I'm not talking about TPUs. I've been programming them myself. I'm talking about the hardware being optimized for fp8, fp4 or just fp2 in some cases. Total garbage except for modeling neutral networks (sans the activation function). And I'm talking about the bonkers rate of depreciation. Locally executed machine learning models? Fine by me if the training isn't setting the planet ablaze & attempts at compensating the diminishing returns by throwing more resources at it.

@mia@movsw.0x0.st @navi@social.vlhl.dev @neal@social.gompa.me @dalias@hachyderm.io @miah@hachyderm.io It's not X.org misbehavior. As a matter of fact: Nothing in X.org itself does talk to GPUs in a way that could stall them. Heck, X.org doesn't even talk to GPUs on that level; what X.org does is associating X11 Drawables with DRM buffers, and the rest is handed off to DRI, having applications talk directly to GPUs. GPU stalls are caused by shader execution not terminating, which is caused by applications submitting shaders that don't terminate.

@mia@movsw.0x0.st @navi@social.vlhl.dev @neal@social.gompa.me @dalias@hachyderm.io @miah@hachyderm.io No. The DDX doesn't work that way. Unless it's running glamor (which it shouldn't because its expensive) the DDX isn't hitting GPU functions at all. Seriously, if you're using glamor, turn that off. You gain nothing from that, except for a lot of wasted memory.

@the_wub@mastodon.social @tokyo_0@mas.to Here's a fun fact: Since Suspend-to-RAM is pretty much miss-and-sometimes-hit on my new Threadripper workstation, I instead went the Hibernate / Suspend-to-Disk route. However, it turns out that saving/restoring system memory state to/from disk takes an order of magnitude longer, than just launching all the programs I tend to keep open from scratch. If there was a coherent method to re-/store session state I'd prefer regular shutdown/poweron over hibernation.

@adriano@lile.cl @SecureOwl@infosec.exchange @acarsdrama@live.acarsdrama.com Deep 😁