"The mistake that every investor, commentator, analyst and member of the media makes about NVIDIA is believing that its sales are an expression of demand for AI compute, when it’s really more of a statement about the availability of debt from banks and private credit."
Soundtrack: Lynyrd Skynyrd — Free Bird
This piece is over 19,000 words, and took me a great deal of writing and research. If you liked it, please subscribe to my premium newsletter.
In programming, we have a nice pair of opposed acronyms:
• DRY for "Don't Repeat Yourself" • WET for its opposite, "Write Everything Twice" (or "We Enjoy Typing")
But there's an intermediate position. The benefit of DRY, other than brevity, is that if a thing is specified just once, the specifications can't get out of sync with each other. If you can't manage that, the next best thing is to make sure the compiler or test suite _checks_ that they're in sync. You have to do more typing than you'd like, but at least you've removed the risk of an accident, which is the _most_ important thing.
(For example, in Rust, if you add a new branch to an enum and forget to update one of its match statements, the compiler complains about the one you missed.)
I feel as if there ought to be a nice intermediate acronym for that state of affairs, so you can say "Weeell, it's not as DRY as I'd like, but at least it's only MOIST." Or DAMP, or HUMID or something.
"Match Or Interpreter Spots Trouble"? "Disallow Almost-Matching Programs"? "Holler Unless Many Instances Dovetail"?
Not sure about any of those. The last one in particular seems especially "you resorted to a thesaurus, didn't you?".
I've got enough pending accessibility-related PRs on my Godot fork such that the editor is almost pleasant to use. Made the scene tree labeled and usable, fixed tab bar navigation so it wraps and doesn't just dump focus wherever if you mistakenly arrow past a tab border, added regions/landmarks to major editor areas so you know where you are, labeled more unlabeled weirdness.
Going to spend the next week mostly hacking on this before switching back to more "productive" work. It might be about time to start working on making the tilemap and audio editors accessible.
I have this crazy idea that it might actually be easier to rewrite and finish System Fault in Godot than to keep up with the perpetual Bevy churn, along with trying to create pathfinding and other systems entirely from scratch and without seeing the results. Even if I rewrote by hand rather than leaning on gen AI, I can drop bunches of buggy pathfinding/visibility code and just focus on the gameplay itself. I'd often said that System Fault could be done already if I could have used an actual, production-ready game engine to build it. I guess soon we'll see.
And to be clear, none of this is a slam on Bevy which has been great. It's just been a huge lift maintaining bunches of code that does what other engines do out of the box, plus keeping up with the inevitable changes caused by building on third-party libraries to make my life easier, then rebuilding when those libraries go away. Bevy will get there eventually, but not on a timeline that has me finishing this thing anytime soon.
I'm done trying to take the hard-line anti-genAI stance that I see so much both here on the fediverse and in the broader discourse among programmers. When I'm truly honest, I have to admit that I find the technology too fascinating, entertaining, and yes, sometimes useful to outright avoid it on ethical grounds. I'm through being a hypocrite. As with my rejection of religion some 14 years ago, it's better to admit I don't really believe, in this case in the arguments against using genAI.
What does this mean specifically? For starters, I won't keep repeating and amplifying the arguments that I've now rejected. I'll use genAI models as I wish, for both business and pleasure. I may post publicly about particularly useful or entertaining things that come out of those sessions, behind this content warning. And I'll start playing around with Claude Code or other similar agents.
But, I'll respect the wishes of any project or future employer that does not want LLM-generated code. I'll review all LLM-generated code that I contribute to any project, and be forthright about the use of genAI in writing that code. And I'll try to be careful about not using an LLM to closely imitate code from known copyleft-licensed projects in whatever domain I'm working in.
Now then, I wonder about the easiest way to run an agent like Claude Code in an isolated, sandboxed *Windows* development environment, as opposed to giving it access to my main box. Just because I'm giving up on strict abstinence, doesn't mean I'm going to be reckless.
I guess I should spin up a new Windows VM in VMware Workstation and install Visual Studio and Claude Code there.
What I have yet to find out is whether I'll *like* relying heavily on an LLM for coding, agentic or otherwise. But I've decided to give it a serious try.
Just as an anecdote on what a top-tier coding agent (Claude Code) is currently capable of, I offer this experiment, where I had Claude Code write a simple custom GUI with accessibility: github.com/mwcampbell/claude-a…
I am, of course, flattered that with no explicit prompting on my part, it decided to use AccessKit. Of course, that's the obvious choice for a GUI implemented in Rust.
The spec for the program is also admittedly very much like the AccessKit winit example, with the notable exception that this program also draws a visual UI.
Considering that, it's a bit surprising that it didn't copy the winit example *more* closely. And as a result, there's at least one bug.
What's most striking as I look through this code is that while Claude decided to use AccessKit for the accessibility part, it implemented simple rectangle drawing/filling and text drawing (with a simple bitmap font) from scratch on top of softbuffer. I told it not to use a full-blown GUI toolkit, but it still could have used vello or tiny_skia or femtovg to handle the graphics, as it used AccessKit for accessibility.
Maybe implementing AT-SPI from scratch is just so obviously complicated that it decided to use AccessKit for that, while it figured doing the graphics from scratch wouldn't be too hard. Or, maybe it pattern-matched my spec against the AccessKit winit example, which has no visual UI, so it improvised on that part. I'll have to look back at the "thinking" traces and see if those give any clues.
I ordered an NVIDIA DGX Spark a few days ago, and it gets here on Monday. It's a very capable computer by any measure, as long as one is OK running Linux on ARM64 as the host OS: 20 ARM64 cores (10 performance cores, 10 efficiency cores), 128 GB of RAM, 4 TB NVMe SSD. *And* it has an NVIDIA GPU (Grace Blackwell 10) with access to the same unified RAM as the CPU. Will be interesting to run some larger local models, and I'll have a powerful Linux box either way.
On the subject of local generative models, the most interesting one so far, in terms of being something that could meaningfully be called "open" (open training process and dataset), is Olmo 3: allenai.org/blog/olmo3 Too bad it tops out at only 32B parameters. The DGX Spark could handle more, though with a shorter context limit (Olmo 3's advertised limit is 64K tokens).
Our new flagship Olmo 3 model family empowers the open source community with not only state-of-the-art open models, but the entire model flow and full traceability back to training data.
Planning to get the same machine for research this year, i'm currently figuring out wether to put the money into this one or a Mac Studio, either of which we are planning to use for AI research. I figure Linux is more flexible, but well... haven't decided yet.
I figure the NVIDIA option is better, if you can deal with Linux (if nothing else you can use that box headless over SSH from another machine), because so much of the AI/ML ecosystem is built on CUDA already.
I've been tempted by kit like this; but decided the expense, space, and energy costs aren't quite worth it to me. I do not use it professionally, so it wouldn't make enough sense. If it becomes more affordable, I'd love to though.
One probably could, in a VM. That box uses UEFI to boot, so it *might* be possible to boot ARM Windows on bare metal, but there would almost certainly be crucial missing drivers.
I'm not sure I'd want to use that box as my main desktop anyway. When it's really crunching away running a big model, it might become too laggy to be pleasant as one's primary desktop. I can just use it as another box on my desk (or maybe on one of the shelves under the desk) that I can access over the network.
From what I hear, if you will be accelerating your AI workloads with an NVIDIA GPU, the Linux drivers perform around 10% faster than their Windows counterparts. Minor, but its still 10%.
OpenAI Codex is much better at Sandboxing than Claude. Not sure how stable its native WIndows sandboxing support is though, and whether you're fine with WSL or whether you need native Windows.
Asking for help! My wife is currently locked out of most digital services due to her phone being broken, and all passwords being in #bitwarden . The bitwarden webapp doesn't help because it asks a verification code, sent on an email which password is.. in bitwarden. What can we do?
- stop hiding which window has focus. It used be done well in 1984 - stop stealing fucking click. If I dismiss one of you fucking dialog the click should not be passed down - start serving user, and I don't mean in a cookbook way
Когда я была ещё маленьким аутёнком и не понимала людей, я решила прочесть одну из книг-инструкций Дейла Карнеги. Прочла. Из всех советов помню единственный: подумали что-то хорошее о человеке — скажите ему.
В подростковом возрасте следовать совету для меня было почти невозможно, но со временем получилось. И всем хорошо: человеку — от комплимента, а тебе — от его радости.
В итоге, понимать людей мне помогли не инструкции, а тонны прочитанной художественной литературы и внимательное наблюдение за окружающими. Так что мало кто разглядит во мне того аутёнка теперь. (А он есть.)
Predsedníctvo Rady a Európsky parlament 16. Decembra 2025 dosiahli predbežnú dohodu o nariadení, ktorým sa vykonáva bilaterálna ochranná doložka dohody o partnerstve medzi EÚ a Mercosurom a dočasnej obchodnej dohody o poľnohospodárskych výrobkoch medzi EÚ a Mercosurom. Nariadenie posilňuje ochranu poľnohospodárov EÚ a zabezpečuje, aby sa ochranné opatrenia mohli uplatňovať rýchlo a účinne v prípade, že dovoz z partnerských krajín Mercosuru spôsobí alebo hrozí, že spôsobí vážnu ujmu.
Predbežná dohoda do značnej miery zachováva rámec navrhovaný Komisiou a zároveň zavádza niekoľko cielených dodatkov, najmä na posilnenie monitorovania trhu a zlepšenie reakcie ochranných opatrení pre citlivé poľnohospodárske produkty. Nariadenie stanovuje, ako môže EÚ dočasne pozastaviť colné preferenciena dovoz poľnohospodárskych produktov z krajín Mercosur, ak tento dovoz poškodzuje výrobcov z EÚ. Stavia na existujúcich ochranných nástrojoch EÚ, ale zavádza rýchlejšie postupy a jednoduchšie spúšťače na ochranu poľnohospodárov v EÚ. Spoluzákonodarcovia sa dohodli na rozšírení zoznamu citlivých poľnohospodárskych produktov, na ktoré sa vzťahuje sprísnené monitorovanie a rýchlejšie ochranné postupy, o citrusové plody.
Spoluzákonodarcovia zaviedli do návrhu Komisie niekoľko zmien s cieľom zabezpečiť, aby EÚ mohla rýchlo reagovať na narušenia trhu spôsobené zvýšeným dovozom poľnohospodárskych produktov z Mercosuru. V prípade citlivých produktov sa za dostatočný dôvod na začatie prešetrovania bude spravidla považovať zníženie ceny o 8 % pri jednotlivých produktoch v spojení s 8 % nárastom preferenčných objemov dovozu v trojročnom priemere alebo 8 % poklesom dovozných cien. Dohoda tiež potvrdzuje navrhovaný rýchly časový rámec prešetrovaní, ktoré má Komisia začať po získaní dostatočných dôkazov. V prípade citlivých výrobkov sa prešetrovania ukončia do štyroch mesiacov a v naliehavých prípadoch možno zaviesť predbežné opatrenia do 21 dní.
Komisia bude neustále a proaktívne monitorovať dovoz identifikovaných citlivých výrobkov a aspoň každých šesť mesiacov bude podávať Parlamentu a Rade správy o vývoji na trhu a akomkoľvek riziku ujmy pre výrobcov z EÚ. Spoluzákonodarcovia sa dohodli na posilnení tohto rámca tým, že umožnia rozšírenie monitorovania na ďalšie výrobky, ktoré nie sú zahrnuté v zozname citlivých výrobkov, ak o to riadne požiada priemysel EÚ. Okrem toho Komisia do 1. marca 2026 vydá technické usmerneniana podporu monitorovania trhu na národnej a miestnej úrovni. Dohodnuté znenie tiež zavádza ustanovenia, ktoré Komisii umožňujú konať v prípade, že zistí obchádzanie ochranných opatrení,a to rozšírením rozsahu pôsobnosti opatrení alebo prijatím iných potrebných vykonávacích opatrení.
Pre bližšie vysvetlenie o čom obchodná dohoda s Mercusorom vlastne je odporúčam video z kanála EU Made Simple. Mimochodom, dosť dobrý kanál, ktorý ľudskou rečou komentuje aktuálne veľké udalosti v Európe a prestavuje možnosti ešte bližšej únie.
Go to https://ground.news/EUMS to stay fully informed on the EU and more with news made simple. Subscribe through my link to save 40% on unlimited access.The...
Poľnohospodárov mám rád a som veľmi vďační za ich prácu. Zároveň však mám v ústach blbú chuť keď vidím aká veľká časť rozpočtu EU je smerovaná na ich podporu - a pritom sa ešte každú chvíľu zúrivo búria 🤬
EU by radšej mala masívne investovať do automatizácie poľnohospodárstva aby sme mali kvalitné, lacné a nízkouhlíkové potraviny, zatiaľ čo mnohí poľnohospodári by sa mohli venovať činnostiam, ktoré viac napĺňajú ich dosť a nevedú ich k rozstrekovaniu hovien naprieč Bruselom...
Go to https://ground.news/EUMS to stay fully informed on the EU and more with news made simple. Subscribe through my link to save 40% on unlimited access.Eur...
Ten pocit, keď SMER hlasujte na každý možný smer, a Hlas hlasuje proti, zatiaľ čo Republika zs 🤔 Takéto kombinácie mám občas v snoch, ale z tých sa zvyknem zobudiť...
Those of you keeping track will know that I did a Christmas market tour early December last year. I ended up downloading the Karlsruhe transit authority’s (KVV) app and buying a day ticket or two off them.
These days I’m more familiar with German transit ticketing, so I haven’t used the KVV app since then.
And today I have received this SMS, telling me they’re going to delete my account soon.
…
⁃ A company that is systematically deleting accounts after a year? ⁃ That takes their data protection responsibilities seriously? ⁃ That proactively informs their data subjects with plenty of time to cope? ⁃ That doesn’t do so using dark marketing patterns that try to guilt me into staying? ⁃ Or upsell me? ⁃ Or trick me into clicking somewhere as the only way to exercise my rights?
I am agog. My flabber is truly ghasted.
Genuinely, this sort of attitude needs to be praised. It’s been 10 years since GDPR, and this sort of thing is still that rare.
⁃ It was sent by SMS to my Spanish number. They don’t care that the number is not German. Their systems will allow this. ⁃ It was sent by SMS, not WhatsApp, which is annoyingly increasingly becoming the default messaging platform for this sort of thing, thus making an American company the mediator of all your messages. ⁃ It was sent in English, my language of choice on their website and app, their processes are capable of checking my language preferences ⁃ No link was included. Nor instructions. If you care about this, it’s obvious what to do. No need to send something that can be turned into a Phishing vector ⁃ My name wasn’t mentioned, no need to attempt to gender me by giving me a title
None of these things are standard yet, and they should be. If you work for a consumer-facing company, can you say that your customer data will be treated this way?
Almost makes me want to keep my account with them, just for the respect they’ve shown
Memphis residents say they're already grappling for clean air after Elon Musk’s AI company began pumping out plumes of pollution to power its gargantuan data center.
You might not notice it in your peer group, when you are not their age, but my daughter showed me that many 20-30yr olds are switching to dumb phones, are cancelling their Facebook/tiktok/twitter accounts and are discovering libraries and printed books. As a conscious decision. Not all of them, but quite a lot and it's growing.
another nice thing you can do with delta chat is having in-chat tools like shopping lists in the family chat where anyone can add or mark an item as done and is marked for everyone, collaborative text pad, to-do lists, split bills, mini-games etc. you can even write your own small utility apps and share it in the chats and your contacts will be able to use them in all platforms
the speed is of chatting is realtime if you use the optimized chatmail relays instead of classic email
Running a whole GUI as root, with a ton of unaudited code running in the same process and that brings in more code across IPC barriers, and has the tendency to poke at random environment variables, mmap random files, and read/write configuration.
Unfortunately the open source ecosystem doesn't have many external signals to users about which components are on the other side of a GUI's d-bus interactions and whether they have been audited. The polkit prompts give many uses the sense that they are granting the GUI application permission to "Run as Administrator" in the language of Windows.
While privilege separation is obviously the right choice, I can kind of understand why people keep on going for the "run desktop app as root" approach: it is a lot easier to do things the wrong way.
If you're building a graphical app that runs as root, it is quite easy to run the app from the build directory without installing it.
To go the privilege separation route, you're going to need to install at least the D-Bus policy files and Polkit action files. It could probably do with some more tutorial style documentation of current best practices here.
it's not like people are going to run GParted sandboxed otherwise. And it fundamentally needs access to your full disk, so not running as root is not going to buy you much security.
But yeah, you don't *have to* run it as root either, so why not run it as $USER.
that XKCD strip is fundamentally flawed: installing fake software running as an admin is how people get access to all those remote services without having to physically steal the laptop.
The problem is not running gparted (or whatever application): it's running everything, from settings to random (GTK) modules, as root without your knowledge or consent. You don't know what else has root access when you run a whole ass GUI application.
I've have bug reports for gdk-pixbuf being broken after the switch to sandboxed loaders because people run gparted as root and modify the environment to set themes and cursors from their user's home directory. This is the kind of insanity we allow, while going around sandboxing and hardening the underlying OS.
IIRC, udisks is pretty complicated to set up (and depends on polkit which can also be quite tricky). Consider that folks who need a GUI want something dead simple. Like “sudo gparted-daemon” and then “gparted-gui” on another terminal.
It also the kind of tool that is usually used quickly in some one-off scenario, not something for which one wants to dedicate lots of time setting up and configuring.
"depends on polkit" yeah, no shit. The privilege escalation flow is a solved problem if you use the tools that the OS provides you. "Something dead simple" is not a thing, unless you want to re-implement everything that already exists and is well integrated with the rest of the OS. In short: people don't want "dead simple" as in "I have to run two binaries"; they want something that works like everything else.
in short: yes, there's some complexity, usually already dealt with by the people developing the applications and integrating your OS. The complexity is not there for shits and giggles, and it's only visible if you're taking over from those two roles above. Some computer touchers enjoy doing the integrator's job, but that's not something that actual users have to care about.
people want “dead simple" and “want something that works like everything else”. That’s why they (try to) run gparted as root.
You can just create a socket for IPC, and use group membership to control permissions. Linux, BSD and any other Unix-like support this just fine.
Perhaps polkit+udisks has some technical advantage (compared to a socket) in some specific use case, but folks will always go for convenience. Especially the kind of folks who want a GUI.
an OpenBSD and Alpine user coming here and arguing for ease of use with handrolled IPC sockets instead of the proper privilege escalation API that is well integrated with desktop environments used by millions of people…
I've been hearing this argument for 10 years, specifically about GParted. What's the reason that the functionality you describe never got implemented? Or was it implemented, and people are just using it wrong?
@Corax42 it's a simple case of nobody putting in the work after the introduction of polkit in 2007; people worked around it with `sudo gparted` because they could.
I have an alarm in my phone called "Germantown." I have no idea why. It was probably a dictation error. But today I need to call someone from Germany at 9 AM, and as I was setting an alarm for it, I realized that "Germantown" is set for 9 AM. So I just turned it on instead of making a new one. Small things amuse me sometimes.
Iran may be on the verge of Internet shutdowns again. If you have friends or family in #Iran, now is the time to let them onboard for example with tiiiz.ir -- because using in-country servers helps to stay connected. If you need support for setting up a relay in or for Iran, send us a DM and we'll see to connect you with other knowledgeable operators. با آرزوی فردایی بهتر...
Please don't overwork yourself. If you feel exhausted, please take a break. Pease don't hesitate to call us via DECT 1023 at any time to sign out of a shift if you can't make it or feel uncomfortable for any reason.
While some people on our team are enjoying the opportunity of learning a lot more things about postgresql, in the meantime here's a rendered version of the map of the event as of the evening of December 24th:
how do you make these decisions around the repo, such as where to host it and what version control system to use? Is it a somewhat rational process or is it a decision from the guts?
its typically a discussion in the team and an attempt to sense what people in general would like and prefer, but ultimately also a good portion of gut feeling...
Můj zlozvyk číslo 1: V rámci úspory času a nachozených kroků jsem začal strategicky přemýšlet, co všechno můžu při jedné cestě z jedné místnosti do druhé udělat nebo vzít. Nejdřív mi to dávalo smysl, ale posléze jsem toto začal praktikovat i v práci, což se trochu začínalo zvrtávat. Typickým příkladem je vaření kafe ráno po příchodu do práce. Do toho si ještě nosím vodu z automatu a někdy i talíř na snídani. Kávovar je ráno potřeba zapnout, vyhodit logr, doplnit vodu. Někdy si během toho ještě strategicky odběhnu na záchod. I když jsem si všechno strategicky plánoval, pak jsem na jednotlivé podúkoly začal zapomínat. Výsledkem bylo, že když jsem se na stejnou trasu vrátil za 20 minut, viděl jsem nachystaný talíř k odnesení, uvařené kafe nebo někde na mě ještě číhala lžička. Cítil jsem se zoufale.
I think one of the wildest parts of the Epstein Files fiasco to me is billionaires are so different that they talk about the darkest most blackmailable secrets of the worlds elite in gmail when I am miserably embarrassed Google is retaining my everyday online cosplay purchase history to train AI.
Die #DB_Bahn sollte sich lieber mal fragen, wieso Menschen mit Rollstuhl 'verladen' werden müssen - wär sie nämlich #barrierefei gäbs das Problem nämlich nicht.
Mal abgesehen davon, dass Menschen mit Hilfebedarf aufgrund fehlender #Barrierefreiheit ja nicht mal spontan reisen können sondern sich vorher erstmal beim Mobilitätservice anmelden müssen.
Der nach meinem letzten Kenntnisstand eine kostenpflichtige Rufnummer hat.
Da fehlt manchmal ein wenig das Feingefühl. Eine ansonsten sehr nette und gut gelaunte Zugbegleiterin am letzten Samstag so "Bitte nutzen Sie die nächste Tür, ich MUSS hier eine Rollstuhlfahrerin rauslassen". Sie hätte zB auch sagen können "leider ist die Bahn immer noch nicht barrierefrei" oder ähnliches. Denn das Hindernis ist ja die Struktur, nicht die Menschen, die genau wie alle anderen einfach Zug fahren möchten.
manchmal? Ein bisschen? Vielleicht solltest du mal @HoernchenCecile folgen...
Und ja klar gibt's nette, freundliche & bemühte Mitarbeitende, das hab ich nicht in Frage gestellt. Die Tatsache, dass diese aber 2025 immer noch notwendig sind, taugt aber halt genauso wenig wie Menschen im Rollstuhl nur als 'Rollstuhl' zu bezeichnen.
Ist ja nicht das erste Mal, dass so eine Durchsage die Runde macht - allein der (mediale) Aufschrei fehlt - im Zug hat's ja auch niemanden interessiert
Die Bahn begründet oft Verspätung mit dem Einladen eines Rollstuhls oder - etwas besser formuliert trotzdem diskriminierend - wegen Unterstützung beim Ein- und Ausstieg.
ich wurde schon mal ausführlich per Lautsprecherdurchsage beschimpft. Bei einem Halt wartete der Mobilitätsservice, ich sagte aber ich wolle nicht aussteigen, ich wolle beim nächsten Halt aussteigen, wegen der Verspätung komme ich so schneller an mein Ziel.
Der Mobilitätsservice verhielt sich als sei es ein "Service" und ich war die böse die absagt. Obwohl es Grundrecht ist mitfahren zu können und den "Service" gibts nur wegen der Barrieren. Aber die Auseiunandersetzung ging hin und her es wurde dirskutiert weil die wollten dass ich da aussteige und nicht beim nächsten Halt und ich was so, ich entscheide selbst.
Dann der Knaller mit der Durchsage als der Zug weiter fuhr. Man habe Verspätung weil ein Rollstuhlfahrer (sic!) - immerhin nicht nur der Rollstuhl - sich geweigert habe ausszusteigen.
Dann kam die Schaffnerin persönlich zu mir um mir auch noch vor den Fahrgästen eine Lektion zu erteilen. Die Fahrgäste solidaridierten sich zum Glück (das ist nicht immer der Fall) mit mir und die Schaffnerin war sauer, ging aber weg.
You can't say you wrote code if you wrote it in C. All you did was write a prompt to a compiler, it did all the hard work of translating that prompt into a string of ones and zeros. Real programmers do that themselves instead of just accepting slow and buggy compiler-generated slop.
Do you know how much water you're wasting by running your LLVM-powered compiler at -o3? You'd be far better off just turning that code into assembly yourself.
Today is not the first time an agentic coding agent has unblocked me. It turns out that the Winlibs package currently ships a version of CMake using libcurl without a valid SSL cert. As such, trying to build anything that requires an HTTPS download will fail, while working everywhere else. This stumped both me and one of my most talented debugger friends yesterday, claude figured it out in 5 minutes this morning.
Ich habe die letzte Woche viel mit #Perplexity experimentiert. jetzt habe ich noch fast ein Jahr einen gratis Pro-Zugang und absolut keinen Bedarf daran. Es gab in der Zeit nur eine einzige Konversation, in der mir kein Fehler aufgefallen ist, der zu Fehlkäufen bzw. Fehlentscheidungen geführt hätte. Was das für die erstellung von Code bedeutet, mag ich mir ehrlich gesagt nicht ausmalen.
Es kann menschliche Sprache glaubwürdig nachbilden. Wenn es um korrekte Informationen geht, selbst im Mainstream, versagt diese Technologie krachend. Ja, vermutlich bleiben LLM. Aber dieser Hype und die Anwendungsfälle, die immer wieder skizziert werden, sind wirklich albern und gefährlich.
Heute: 13 km radeln mit der Flims (einer der vielen Ylvi-Spitznamen), dann Kamin, Kaffee und Kuchen zu Hause und gleich noch ne große Runde mit den Kleinis und dem Mann. Zu Hause isses halt doch am Schönsten. 😅
I did a quick unscientific comparison. AI generated drums and bass for a snippet of a cover version versus me hobbling through those parts with software instruments, playing from memory of hearing it recently.
I invested a similar amount of time in total to reach the second verse of the song. With the generative engine, I only started paying attention to time spent *after* figuring out enough of the horrible UI accessibility to be able to get anything done with it, so although that will inevitably cloud my judgment of whether I enjoyed the UX, it shouldn't impact the comparison of musicality/usefulness.
In both cases, the original song was playing for reference with its rhythm section removed. I've turned the original down pretty quiet in these examples, loud enough to be recognisable but hopefully balanced in a way that makes it easy to focus on the differences in drums and bass.
The generative engine had this prompt: "Follow the chords and dynamics of the reference track as closely as you can. I'd like a laid back feel, fingerstyle bass playing and a 70's style deadened drum sound".
This is as good as I could get from it after a bunch more prompts trying to help it hone in. 2/4
I played the drums and bass in this version on keys here, using software instruments, so a totally silent recording setup that I could make at 4AM with a sleeping partner in the same room. I'm not much of a keys player and I don't have absolute pitch or a good enough memory to recall parts note for note, hit for hit, this was just a gut instinct jam-along.
Drums were recorded in two passes, I played the kicks snares and toms first, then overdubbed hats and cymbals. In both cases I did some light tweaking of the MIDI to get a few stray velocities and timing gufties straightened out. 3/4
@KaraLG84 I thought it would be better. When I heard the first few notes of the bass track soloed my initial reaction was like "hey, that's a decent thumpy bass tone, it's playing the sort of pattern I'd expect and... oh..." then the rest happened lol.
Why are you using generative ai when you have focused so much of your time on making reaper accessible and do your own music? Also, now that you have uploaded music that you made to I am guessing suno or udio, now they have your music and can train there ai off of what you made. You basically just gave them free work and they don't have to pay you for your time and they can just steal your music so someone else can make money off of it.
@michaeljohann_ That wasn't my music, it was a cover of a hugely established band that will have already been scraped for training if that's how they're doing it. This was a quick experiment to see what generative AI provides in this context. As someone who teaches a lot of folks coming into making music for the first time it makes sense for me to be at least broadly informed about what's out there.
Fare enough as a musician I really don't like to see people using there own music for AI for the reasons I mentioned earlier I understand wanting to learn it to educate people about it. I just tell people to stay away from it and spend the time you would to use ai to learn a daw or an instrument. You will get better results and actually learn something rather then typing in prompts to a computer, but maybe I shouldn't be so harsh.
@michaeljohann_ I'm advising the same in a lot of cases. Thing is, there are an increasing amount of times when people don't listen to that advice, so, figure I may as well know what they'll find.
@chikim There wasn't any mixing. I did a small amount of tweaking of the starting presets in each instrument just so they sounded a bit closer to what I was imagining, nothing drastic enough to make or break whether they'd sound workable though. Then I turned the volume of the original down before I rendered, which wouldn't have been a necessary step at all outside of this comparison context. That's all. Do those choices need democratising? Who for?
@chikim In this case, they'd come away with a quick demo that didn't follow 4/4 timing or simple chords. I'm trying to think of who that's useful to, and I'm coming up empty.
@chikim Well sure, I understand what democratizing is and in many cases I'm in favour of it. TBH I can't really say I'm against it in this scenario either. It's more like... I dunno... I'm surprised that even as a barely function keys player and probably less than 5 minutes of tweaking sounds, I could throw together something that's more usable than this thing that some people are paying money for.
@chikim Last week I had a session helping someone remake a beat that was from Suno. I've heard plenty of examples now where the generative part coming first then humans doing their thing after/on it can turn out good. In part that's why I'm surprised this gave me such a shit result. I didn't think the order of events being flipped around would make the result so different.
@chikim This flow works for me. More than a few times I have heard an AI version of something and went "hey, that was fun, I bet it'd sound awesome with real people." The AI stole everything from us, we can steal it right back.
Was ich mir fürs nächste Jahr von Männern wünsche: Schweigt nicht bei sexistischem Verhalten anderer Männer. Schaut nicht betreten weg. Sagt etwas. Macht etwas. Das ist auch Eure Baustelle.
Wenn ihr Teil einer Gruppe seid, bei der sich jemand sexistisch verhält und ihr lasst es geschehen, weil: „Ist ja nicht meine Sache, muss er selbst wissen, usw…“, dann seid ihr aus Perspektive der Betroffenen nicht „Einer von den Guten, die aber nix sagen“ - ihr seid Teil des Problems.
**altbot v2.5 is out!**
fixed non-english support! switched to a two model setup: gemma3:1b handles translations while qwen3-vl does the image processing stuff. gemma3:1b is surprisingly good at translations for such a tiny model??
now running 23.
I am blind. Seeing people who think I'm not worth the effort fill my timeline with AltBot generated AI stuff that isn't even accurate in lots of cases.
Human alt text is always better, because it doesn´t focus on ocular seeing. Seeing people think, and AltBot was designed around that notion, that blind people must compensate for missing "eye-seeing", but that's not the case. I am interested in the meaning of an image to you, its maker or publisher.
Again, human alt text is better, also because it strengthens reciprocity between seeing and blind people. AltBot doesn't but it makes seeing people believe they have done their bit for accessibility. In actuality, the reverse is often true.
Lavrov: Západ musí chápat, že Rusko má na Ukrajině strategickou iniciativu
(ČTK)
https://www.ceskenoviny.cz/zpravy/lavrov-zapad-musi-chapat-ze-rusko-ma-na-ukrajine-strategickou-iniciativu/2765595
What's the good in having this apartment with it's fancy amenities (pool, barbecue) if nobody ever shows up to visit? They all said oh yeah we'll come see your place. Half of them haven't. Half of them aren't getting the fucking hint when I post on facebook. So I'm done.
Wanna swap? This is a once in a lifetime opportunity to live in a damp, cramped shit-tip that benefits from noisy neighbours in 3 directions and a wonderfully hassle-free existence by foregoing pools, barbecues or any attraction that might tempt people to visit. Act now to lock in the joys of blocked drains and a stink of sewage that can be experienced in the overgrown garden.
@Scott Awesome. We'll swap. You can have my crackhead parade, screaming children and barking dogs because people leave them home for most of the day, plus cafes and train tracks that are in bad need of work. Think we're getting good deals here haha.
@sapphireangel @Scott My place is positively utopian based on everything previously. I live in a Cul De Sac, we tend to hear trains every few minutes but they're not obnoxious, got our own back garden which is bigger than a postage stamp which, considering it's London has to be some kind of miracle, a real piano which I can play whenever and no neighbours complain. The downstairs toilet door handle needs changing but that's about it. lol
@FreakyFwoof @sapphireangel @Scott haha Knightly the piano, gotta love the sound of that thing. Here it's nice and quiet in Banesberie Close Banbury as it's mainly elderly that live here, the back garden's nice and straight and big, kept neat and tidy by Edward my carer and sometimes a gardener to help. No crack dens here. Nice and quiet.
@TomGrant91 @FreakyFwoof @Scott My apt is great for me. The porch is too small though, but I make do. The yard is often in need of mowing, but I get my peace and quiet. It is currently icy as hell though.
@sapphireangel @TomGrant91 @FreakyFwoof @Scott My flat is slightly noisy with traffic but am used to it now. Fairly hot especially in the summer but has a balcony and access to a well maintained shared outdoors space with children playing. Family want me to move to a larger place but right now don't think that's happening in London as I'm about a minute walk from the tube.
@andrew I haven't forgotten about going for that curry as well. I think we've found the cheapest place in London but with the very best food around. Went there in December with my mother in-law and sister in-law, along with Jake. For four people, three of us paid £38 each. Definitely not a bad run. Food was excellent, everybody came away very happy.
Best Indian Restaurant & Takeaway in Brent, London. Order authentic Indian food, covering Cricklewood, Dollis Hill, Willesden Green, Kensal Rise and nearby areas.
@FreakyFwoof Right its looking like Monday evening may be an issue now so preferably Friday or Saturday. I can also do Sunday evening but apreciate that may not be good as its a school night.
@andrew OK well we'll aim for a free Friday then, as there's no rush to get home for a particular time. Think it's probably easier. I mean I don't picture a midnight food run or anything but it does mean a bit of extra sleep next day after filling the belly.
The amount of discussions I have to hold because #Laravel devs don't want to write a read-only DTO somewhere, but instead add a dynamic property on some ActiveRecord contraption.
It's so exhausting, and it could be so simple, if it wasn't that the framework promotes terrifyingly bad software practices everywhere. #php
I took my final decision not to use Laravel unless absolutely needed (like paid for this) when I added lots of typehints, sent quite a large PR, and it was declined because "it's not needed at the moment". Then I said no, sorry not sorry. I like PHP's evolution, and writing in #PHP 5 style is something that doesn't inspire me.
Kaden
in reply to GNOME • • •GNOME
in reply to Kaden • • •