Mastodon Feed: Posts

Boosted by pixxl:
krig@goto.liten.app wrote:

@Hypfer I think saying that's a neutral amplifier is oversimplification, that's not what it does. There is a statistical inference happening based on the training data.

Imagine that it is trying to finish a sentence for you. You are not simply getting "more sentence", you are getting a somewhat statistically likely average of what the rest of the sentence might look like. So you what you get is likely to "make sense", but be flat. It won't generate anything brilliant or come up with something that no one has seen before (other than through juxtaposition), it can't. So it's not just giving you more or amplifying what you put in. There is a direction to it.

It's easier to see with images and designs than with code. All generated images look the same. They have a lifeless, copycat quality to them, no matter how good they get. All of the AI-generated websites look the same, and they all look like shit. Every startup looks like shit now. And I really think it's bad for people to interact with this technology because while you may think that you are the one in control, the LLM output is training you too and you are adapting to it as much as it is doing what you tell it to do.

If you spend your days interacting with LLMs, you will take psychic damage from that. I am convinced of that. Part of that damage, I think, is losing the ability to see just how shit it is. So when you're feeling productive creating all these things, you are not seeing the sloppiness of what it's putting out, how shit it is. And aren't you robbing yourself? Would it not be more worthwhile to make just one thing that is uniquely yours, that you can be proud of, than all those things that you aren't showing anyone else, that you didn't make, that you don't value and that ultimately just degrades you? I mean, I really think so. I think you are making yourself less by using these tools to replace that part of you which is uniquely yours. Your ability to make things that no one else could make.

Boosted by pixxl:
krig@goto.liten.app wrote:

@Hypfer Don't you think it is unethical to use technology that has such enormous externalities to do things that you value so little that if you had to do them yourself you wouldn't? But I also keep hearing people say that it is making them productive. Yet I see nothing of value being produced, by those people or in general. Is it possible that the perception of productivity that you are experiencing is not actually backed by real productivity?

Boosted by pixxl:
krig@goto.liten.app wrote:

Maybe I am the weird one. I am definitely the weird one, OK? I've managed to get this far OK in life, but... waves hands around, exasperated This whole thing has never made sense to me.

Maybe that was why I was drawn to black metal and outsider art as a kid. Maybe that was why I left my home town as soon as I could and return as seldom as I can.

Anyway.

I happened read this article by Martin Fowler: https://martinfowler.com/articles/2026-dont-like-llms.html ...not because LLMs interest me, but because people interest me, they fascinate and confuse me.

Martin Fowler confuses me. I fundamentally don't understand how he thinks, how he can reach the conclusions he does. In the article, every reason and hope he has for positive outcomes from using LLMs are either fear-based or unsubstantiated bullshit. Every fear or worry he has is over something that has already come to pass.

"excited by the potential gains in productivity"

Maybe this was a view that it was possible to hold in 2023? By now it should be obvious that volume of output does not equal productivity. I mean it has always been obvious. We ALL KNEW that counting lines of code was a BAD FUCKING MEASUREMENT and yet you massive idiots can't stop counting lines of code as if you're convinced that God will judge you based on the amount of shit code you produced while alive.

Just look around at the state of software and the world. Where is the productivity, and is it doing us any good?

"I’m fearful of the damage AI might cause: agent swarms taking over our virtual and physical infrastructure, designing bio weapons."

The first of these has already happened! You don't need to fear what is already here. And why limit yourself to bio weapons? AI is already being used to murder innocent people in Gaza and Iran to name just two REAL PLACES where REAL PEOPLE LIVE and where LLM technology is being used to kill children, to target innocent people for crimes they haven't committed, to lie and give false testimony in court, to destroy the social fabric of trust and the democratic process.

Everything you fear is already real! Everything you hope for is a lie!

How can someone exist in the world and not see this? I don't get it!

"One of my most successful life-hacks is to avoid people I don’t like or don’t trust. I decline to interact with them socially, and make a deliberate effort to avoid working with them too, even if they are doing much that is beneficial. I feel that hanging out with pleasant, capable people, the people with integrity, has made my life a far better one. Hence my visceral dislike of interacting with an LLM that’s not just making a pretense of being human, but also posing as the kind of human I walk away from."

So just fucking don't! Like what the fuck dude.

Boosted by pixxl:
krig@goto.liten.app wrote:

"Right now, AI is mostly used for frauds, propaganda, misinformation, generating child porn, scams, spam and warfare. Also the companies producing it thinks there's a 10% chance the technology will destroy humanity. Also the AI is an annoying piece of shit that I hate to interact with. IN CONCLUSION WE HAVE A RESPONSIBILITY TO USE IT BECAUSE PRODUCTIVITY"

Boosted by pixxl:
liw@toot.liw.fi ("Lars Wirzenius") wrote:

It is with a heavy heart I need to remind everyone again that the only true and valid definition of productivity is how much happiness and well-being is being produced.

slightlyoff@toot.cafe ("Alex Russell") wrote:

The Intel NUC 13 Extreme (pour one out) has been an incredible, quiet gaming rig for a few years; particularly after replacing stock fans for Noctuas. But one component has created intermittent instability: the PCH chip. Last year I tried to "fix" the non-contactng heat sink with thermal putty, but still saw high temps under heavy load, and after a crash in Zero Company, I'm abandoning half measures.

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

please, please, please refrain from replying to me asking "did you expect anything different" from meta. obviously i expect it to be a big ball of surveillance and trash. that is why i am looking at it. me describing something does not mean that i am surprised by it unless i say something like "i am surprised by this."

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
carnage4life@mas.to ("Dare Obasanjo") wrote:

All of these parents should face charges for negligence. Every single one.

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
RunRichRun ("Rich Stein (he/him)") wrote:

@mattblaze
Of course, you know where computer "bugs" come from — https://www.computerhistory.org/tdih/september/9/

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
Ivor@mstdn.social wrote:

Human: "I'll just put this box upside down on the stairs so the cats don't jump into it".

Cat:

#Caturday #late

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
dansup ("Daniel Supernault") wrote:

Jake Signal Strength: 100%

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
cdarwin@c.im ("Chuck Darwin") wrote:

NOW
When TechCrunch reached out to Meta on Saturday, a spokesperson said, “This was an error and the ads are being restored.”

We’ve asked Meta for more details and will update this article if we hear back.

YouTube also posted an update Saturday, noting that “the trailer has always been available on YouTube.” (Indeed, that’s how we were able embed it at the end of this article.)

The company added, “When submitted as an ad, it was temporarily restricted by our system. Following review the ad has been cleared to run.”

X’s rejection is unsurprising
— Musk, who owns the social media platform, recently posted that “Dogshit is worth more respect than Gibney"

https://techcrunch.com/2026/09/26/meta-says-it-will-run-ads-for-musk-documentary-after-all/

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
DamonCrowley@mindly.social wrote:

🚨A Bible-thumping North Carolina Trump supporter who called for public hanging of p*dophiles, Shaun Michael Anthony Frank, 23, has been charged with statutory r*pe of a child 15 or younger and indecent liberties with a child, tied to an alleged 2024 assault on a 15-year-old girl.

He also faces a second-degree forcible r*pe charge connected to a 2025 incident involving another woman who was reportedly physically helpless at the time.

Frank is being held in the Onslow County Jail without bond.

Before the arrest, Frank’s Facebook was a steady stream of Christian “family values” content. He shared MAGA memes thanking Trump for “fighting for the American people,” pushed anti-trans posts insisting science only recognizes two genders, shared anti-Planned Parenthood propaganda, and reposted Christian nationalist “rapture is near” content tied to Israel.

But the post that hits different now is one from 2019, where Frank shared a meme declaring he supports publicly hanging p*dophiles and s*xual predators — right around the same years prosecutors say he was committing the crimes he’s now charged with.

This is the pattern RWW keeps documenting: the loudest voices demanding brutal punishment for predators are too often hiding behind that faux outrage.

Not a drag queen. Not trans. Not undocumented. Not a Muslim. Just the usual suspect.

ireneista@irenes.space ("Irenes (many)") wrote:

(but we just want it more declarative)

ireneista@irenes.space ("Irenes (many)") wrote:

(if you wanted to compile something other than the compiler, up to now the recommended way would have been to copy evoke.e, which is nice and short, and modify it appropriately)

Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
digyoursoul@universeodon.com ("Voting is Your POWER") wrote:

Archaeologists have found a Chimu royal tomb in Peru that centuries of looters somehow missed.

Under the ruined citadel of Chan Chan, once the capital of the Chimu kingdom and the largest city in the pre-Columbian Americas, a Peruvian team cleared 40 centimetres of sand from a doorway before their flashlights found funerary bundles still undisturbed after 600 years.

Inside the mausoleum, 38 bodies, some with red cinnabar pigment on the skull, alongside silver and copper jewellery, quartz beads and a wooden chest of textiles showing seabirds and monkeys.

The tomb’s design suggests it was built to be reopened again and again, bodies carried out to be venerated in the city’s plazas before being returned.

https://english.elpais.com/culture/2026-09-10/an-intact-tomb-preserved-for-600-years-opens-a-window-onto-the-world-of-the-chimu-elite-in-peru.html

jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:

nice rant

https://modernmotherfuckingwebsite.dreamstation.systems/

Boosted by pixxl:
jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

Mfs are so broken by the prestige treadmill they've forgotten the simple joy of "having an idea"

Boosted by pixxl:
jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

I have heard people talking about submitting tons of AI grants they don't even really know the full details of because "something might eventually get through" and I have to look away to avoid being drawn in to the vacuousness of their being

Boosted by pixxl:
jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

I wonder what the effect size of "believing academic writing is just a mechanical paint by numbers factory process" -> "writing with AI" is. Academic writing is a genre and a literary form, and like all genres and literary forms there are idioms, expectations to play with and violate, a range of expressive devices you can draw on or discard. People who don't treat academic writing as a craft or an art write the worst papers.

Boosted by pixxl:
jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

Its good to know how to write, actually, and its very hard to learn, and the only way to learn is to do it a lot.

Boosted by pixxl:
jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

RE: https://neuromatch.social/@jonny/117164185311151783

A friend who just won a big grant just said to me "I think not using AI might have made my grant more competitive than it would have been years ago" and like... As long as there are people who actually read the grants, it gives me a schadenfreude laugh to think of the prestige class pumping out what they think is 10x solid gold genius, but to everyone else it just looks like they are surrounded by a cloud of pink slime like pigpen in peanuts.

Boosted by pixxl:
david_chisnall@infosec.exchange ("David Chisnall (*Now with 50% more sarcasm!*)") wrote:

The most toxic thing to happen in the F/OSS world is the attitude that some people are so smart (because they’ve created successful projects) that you cannot criticise them. The idea that, because someone created something good, they are infallible is disastrous for them, the projects they run, and the broader ecosystem.

Producing good work is inseparable from getting useful feedback.

I like to think I’m pretty smart, but I absolutely would not have been able to build any of the things I’m proud of without people willing to tell me that my ideas are stupid.

There are may examples, but a couple of concrete ones: I had an entire mechanism designed for delegating access to the CSR field that allows control over interrupts in CHERIoT. Hongyan Xia pointed out that I was solving the wrong problem, badly. He proposed sentries (an existing mechanism for immutable, unforgeable, function pointers and return addresses) that enabled or disabled interrupts on a jump and captured the prior state in the link register. This not only let us have functions that ran with interrupts disabled (and therefore the ability to reason about worst-case context switch time), it also let us statically audit these properties and enforced a structured-programming discipline. Murali Vijayaraghavan later extended the mechanism to properly differentiate function pointers and return addresses, giving lightweight enforcement of some useful CFI properties. I’d not tried this because I assumed it was impossible without explicit call and return instructions, he showed me I was wrong.

If I ever reach the position where people are unwilling to disagree with me, I expect the quality of my work will drop considerably.

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

ope i have a submission for the PII leak bug bounty lmao

Boosted by ireneista@irenes.space ("Irenes (many)"):
xgranade@wandering.shop ("Cassandra has had an adventure") wrote:

Software developers please understand both copyright law *and* labor solidarity.

Boosted by jwz:
ComicContext@mstdn.social ("Comics Outta Context") wrote:

Boosted by pixxl:
zip@furry.engineer ("Zip") wrote:

@terrance_shaw
I'd rather get fired than sell my soul to the machine
@soatok

ireneista@irenes.space ("Irenes (many)") wrote:

we've probably got the first part working now, and we're just starting on the second

ireneista@irenes.space ("Irenes (many)") wrote:

so today's effort on Evocation is to move most of the code that writes an executable into a helper (we're calling it "enchant")

and then move the code that writes the hex dump of that executable into another helper ("inscribe")

deceptively fiddly, though it's not actually a lot of work. the hope is to turn it into reusable machinery that can compile things other than the compiler itself.

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

So zuck has mentioned this concept of "fleet learning" in interviews, and here it is. Strings extracted from hatch describe "fleet learnings" - or, things that other agents have learned and submitted to a "fleet learnings exchange." agents are instructed to derive generalizable ideas that can help other agents with their various tasks - which, again, include mining someone's personal life for information about their relationships and shopping habits, and generally deepening dependency on the tool. Zuck described this feature as being used only for "ideas" - tasks presented to the user in the ideas panel - but it is invoked more broadly, e.g. in the "alignment synthesis" job, which is the nightly task that determines how to "strengthen your relationship with this user" (first and second pic are from the same instruction, "consult the fleet for this task" where the task refers to alignment, of which second pic is a subset)

Note the conflicting instructions: "de-identified, not denatured" - the LLM is supposed to strip out personal information, but make sure they enough of the lesson survives that it's useful. So in the best case where PII removal via LLM judgement works perfectly (it won't), muse agents participate in a horizontally-exchanging swarm where they learn how best to manipulate people by their personality traits and behaviors. The dumbest distillation of the targeted advertising economy you could ask for.