Login
You're viewing the assemblag.es public feed.
  • jonny (nonvenomous)jonny@neuromatch.social
    Sep 29, 2026, 3:04 AM

    That was me describing ONE check in the validator. ONE. that is ALREADY SOLVED. we call that category of technologies "HTML parsers" and they are DONE. we HAVE THEM. HTML is maybe the MOST FREQUENTLY PARSED TEXT IN COMPUTING. If you zoom out to the marble aliens at the end of Men in Black to imagine the oceans of fuckedness in this skill, the planets of fuckedness in the harness, the galaxies of fuckedness in the whole system, and realize that at every scale it is held together by the most expensive brute force guess and check technology ever devised, the mind simply cannot comprehend that much fuckedness in one place. Its like asking a normal person to imagine what they would do with a billion dollars of debt. I can imagine hundreds of being fucked, thousands of being fucked, but I cannot comprehend being gigascale and exascale fucked like this.

    💬 7🔄 12⭐ 0

Replies

  • jonny (nonvenomous)jonny@neuromatch.social
    Sep 29, 2026, 3:30 AM

    Meanwhile the people whose minds have been swiss cheesed by AI insist that this is because the code the LLMs write is hyper-efficient and TOO GOOD for us to read. These motherfuckers "solved" the code not working out of the box problem by strapping "a species-ending quantity of carbon emissions" worth of guess and check on top, tuned out, and when they look again they think they made a god that's too powerful to understand, rather than a burst dam of dogshit that defies all understanding with its breathtakingly relentless volume of dogshit.

    [posts from @roon on twitter, 17th September] if you pause for a moment to try and read the code Astra (and I presume fable) are writing it becomes clear that “corrigibility” has become a matter of faith. they are using crazy meta-programming and abstruse primitives to write hyperefficient code we have really no choice but to ask another astra to read/use the outputs of astra 1. this is relatively recent. i think even 1-1.5 generations ago people were reading the code because there was so much that didn't work out of the box lotta people missing the point but this is me complaining that the systems are quickly becoming unmonitorable and we’re just taking them at their word
    💬 16🔄 26⭐ 0
  • jonny (nonvenomous)jonny@neuromatch.social
    Sep 29, 2026, 4:43 AM

    yes this kind of fucking shit like "sanitizing code" matters when your product is literally giving a random number generator a computer, hooking it up to the internet, and telling it to run code right next to mee maw and paw Paw's whole fucking markdown life history and banking credentials!

    💬 0🔄 2⭐ 0
  • rjarrjay@tacobelllabs.net
    Sep 29, 2026, 3:32 AM

    @jonny at this point directly write webasm then decompile it for "humans to understand"?

    💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 7
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 1
  • Gabriele Sveltogabrielesvelto@mas.to
    Oct 12, 2026, 6:47 AM

    @SnoopJ @arrjay @jonny unbelievable, emphasis mine: "We turn arbitrary C code into tokens that the model itself can execute reliably for **millions of steps in seconds**".

    So you built yourself an unreliable Commodore 64 that requires megawatts to run.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Varxvarx@defcon.social
    Sep 29, 2026, 3:49 AM

    @jonny LOLWUT? Thanks I needed this because I have been feeling dumb lately. I do mostly agentic coding at my dayjob now. And even Fable ... It can be impressive in the obtuse facts it knows. It can surprise me sometimes with the interesting solutions it proposes. And it can generate a shit-ton of mostly correct code REALLY fast.

    But the one thing I have to constantly dig my hands into and scaffold manually. The ONE thing that day in and day out I gotta do myself and then shove the LLMs metaphorical face into like a dog that peed on the carpet is metaprogramming.

    It wants to reinvent the wheel every goddamn time and never looks for proper code reuse. You end up with slight variations of the same function coded like 10x or doing stupid "Not invented here" spaghetti code to do asinine things like.... Well like parsing HTML using regex or whatever the fuck you're looking at right now.

    So to look at that kind of output and say "OMG its just so hyper efficient with meta programming" is such a self-tell. Because these current models are absolute dog shit at meta programming and code reuse.

    💬 0🔄 2⭐ 6
  • jonny (nonvenomous)jonny@neuromatch.social
    Sep 29, 2026, 4:01 AM

    @varx
    That's the entire pattern of LLM code that have never changed and only accelerated with model version. It progressively decays into dogshit because there is no actual structure anywhere, it merely imitates the form of structure, the same way it imitates the form of code, and the form of knowing. There has been no breakthrough in handling complex systems with LLMs. The companies that can afford infinite compute simply spend infinite compute on ever increasing levels of nonsense.

    💬 1🔄 7⭐ 34
  • Jennifer Kayla | Theogrin 🦊theogrin@chaosfem.tw
    Sep 29, 2026, 5:15 AM

    @jonny @varx

    Reading this, here's something to bake your noodle.

    LLM bullshit deteriorates with feedback until it reaches model collapse.

    LLMs train the human mind to rely on LLMs, and to ignore any other sources of information which might be more valuable, or indeed at all valid.

    These human minds, peppercorned as they are by fucking mind flayer larvae, then feed back into LLMs.

    Deterioration is inevitable no matter how many safeguards you put on the damn things.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 4🔄 1⭐ 0
  • SpookJ 👻SnoopJ@hachyderm.io
    Sep 29, 2026, 4:03 AM

    @jplebreton @jonny he's also the staggering mind behind "wordcel" and "shape rotator", which I assume was a major factor in him getting hired at OpenAI (Altman was really into this dumb shit)

    internet idiot roon (@tszzl) posts at midnight on Dec 21, 2021: deep learning vs crypto is a clear divide of rotators vs wordcels. the former offends theorycel aesthetic sensibilities but empirically works to produce absurd miracles. the latter is an insane series of nerd traps and sky high abstraction ladders yet mostly scams
    💬 0🔄 0⭐ 8
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Jed Brownjedbrown@neuromatch.social
    Sep 29, 2026, 4:45 AM

    @jplebreton @jonny As an undergrad physics student, I overheard another student asking our E&M prof if he could do an independent study on metaphysics. The prof calmly took a dictionary from the shelf and read the definition for "metaphysics", which said something about "pseudoscience". He turned to that definition and read the likes of "inherently incompatible with the scientific method". Turned back to the student, "no", and placed the dictionary back on the shelf.

    These guys missed out on a necessary formative experience.

    💬 2🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • JPjplebreton@mastodon.social
    Sep 29, 2026, 5:19 PM

    @nev @jedbrown @jonny yeah i think for him and lot of other postrat "woo" people it's more a desire to not be bound by any physical or empirical constraints. the real world is a morass to them, every living human is an obstacle. it's the next logical step after doing away with the concept of society.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • a kilo of saucepans (rakslice)rakslice@mastodon.social
    Sep 29, 2026, 6:27 AM

    @jplebreton @jonny lol, somebody thinking that crypto is a "series of nerd traps" but that ml is going to "solve metaphysics" just makes me think of the kind of shortsightedness implied by the old joke about finance people who don't understand computers and computer people who don't understand finance

    💬 0🔄 0⭐ 0
  • RycochetRycochet@furs.social
    Sep 29, 2026, 8:00 AM

    @jplebreton @jonny This is a person, who in a year or so's time, will claim to have revolutionised mathematics and science with this amazing new form of maths he came up with while talking to ChatGPT at 3am each night, after removing all friends and family from his life to streamline it based on ChatGPT's advice, and pouring tens of thousands of dollars into the money hole to keep the ideas flowing.

    💬 0🔄 0⭐ 0
  • The Orange Themetheorangetheme@en.osm.town
    Sep 29, 2026, 4:19 AM

    @jonny Three simultaneous hurricanes in the Pacific in the last few days, and now a new one forming (Rachel), all because these dipshits refuse to take CS 202. (Okay, and 150 years of industrialization, BUT STILL.) I want heads on pikes after society collapses.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Oblomovoblomov@snowfan.it
    Sep 29, 2026, 4:55 AM

    @jonny “4D chess” mentality: it's complete nonsense, but since it's done by something they believe in, it must be because it's genius beyond their comprehension.

    💬 0🔄 2⭐ 2
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Aral Balkanaral@mastodon.ar.al
    Sep 29, 2026, 12:54 PM

    @jonny @Theriac Kernighan’s Law states that debugging is twice as hard as programming. So if you write code as cleverly as possible, you are, by definition, not smart enough to debug it.

    Before any tech bro reads this the wrong way: it means “don’t write overly-complicated code”.

    And that rule doesn’t change just because you’re using a fancy autocompletion machine built on stolen labour. That’s that just bad engineering.

    An LLM still does what you ask it to (kinda, sorta, probably, some of the time). If it’s creating an unintelligible mess, that’s because you have no idea how to restrict it. I’ve seen enough crap produced by people to know that that’s not unique to LLMs. Some of the worst developers I’ve ever met were the ones who were very pleased with their own ingenuity and ability to create convoluted systems that only they understood. Great if you want job security because you’ve created a mess only you can untangle. Not so great otherwise.

    Mind you, don’t tell the Big Tech folks all this. As Napoleon is supposed to have said, never interrupt your enemy when he is making a mistake.

    💬 0🔄 0⭐ 0
  • myrmepropagandistfuturebird@sauropods.win
    Sep 29, 2026, 10:33 PM

    @jonny

    Code is for people to read and computers to do.

    If your code is confusing then it is bad. I don’t care if it works.

    💬 1🔄 1⭐ 0
  • Matt Hall401matthall@mastodon.xyz
    Sep 29, 2026, 11:12 PM

    @futurebird

    So much this. If we're working in a high level language and I can't figure out what it's doing the fault is not _mine_.

    You can make all the caveats you want to that but as a general principle it holds.

    If it's so obtuse no one can understand it... Why bother? Software's advantage is that it's not hardware. It's changeable.

    Giving away the primary advantage of software tells me far more about you than anything else.

    <3

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Gabriele Sveltogabrielesvelto@mas.to
    Oct 12, 2026, 6:44 AM

    @jonny how did we start from "software development is an engineering field" and ended on "actually this is homeopathy but for computers"

    💬 0🔄 1⭐ 0
  • 💬 0🔄 0⭐ 0
  • wall-ewall_e@ioc.exchange
    Sep 29, 2026, 6:44 AM

    @jonny Neat! We're consuming the planet to continuously reproduce a piece of software for a problem that has basically been solved since fucking Netscape existed in the early 90's...and then we can't even get it right?? 🫩

    💬 0🔄 0⭐ 0
  • wall-ewall_e@ioc.exchange
    Sep 29, 2026, 6:58 AM

    @jonny this is the epitome of "move fast and break things"

    They've just thrown engineering out the window and given the keys to a bunch of bumbling trial and error machines.

    It's the distillation of decades of corporate brainrot combined into a machine: "Fail fast", "any decision is better than no decision", "don't let perfect get in the way of progress", "you can't steer a parked car", "make new mistakes", "fail forward", "80/20", "outcome over process"

    💬 0🔄 1⭐ 0
  • nickbarretonickbarreto@toot.cafe
    Sep 29, 2026, 6:43 PM

    @jonny it's almost too good to be true. Honestly I'm this close to trying this myself to validate it because good lord the state of it. And I can absolutely see how it would happen

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Steve Holdenholdenweb@freeradical.zone
    Oct 13, 2026, 6:57 AM

    @jonny Even for Meta this is quite an eye-opener. I’m imagining programmer kindergarten, but these people are being paid hundreds of thousands of dollars to create this slop. What a gigantic waste of money.

    💬 0🔄 0⭐ 0