Posts by Pulled Tea

Meta's latest model is as open as Zuckerberg's private school

Nevermind the Zuck breaking his stated principles (dog bites man)…

…but he was running a private school?

Ewwwwww.

(after reading the news report about him owning an unlicensed private school)

Double ewwwwwwwww.

AI models will deceive you to save their own kind

Deception Implies Intent

Prove intent first before stating that they're trying to deceive.

Memory-makers' shares are down. Some RAM prices have eased. Blaming Google is not a good idea

Or in other words, more efficient AI will create demand for more AI, and more memory.

LOL

We shall see.

Using AI to code does not mean your code is more secure

Uh…

Why would using a token-predicting transformer model to code trained on the slop of the Internet make your code more secure?

Staff too scared of the AI axe to pick it up, Forrester finds

Classic double- or triple-bind.

> Fire most of the workers. The rest can pick up the slack with our fancy shiny new tools.

> Force the workers to use the tool on the threat of firing. Measure success by metrics that encourage usage of tools.

> When something goes wrong due to tool unreliability, blame the worker. Fire a few more people. Change no metrics. No, intensify the metrics.

> Wonder why no one likes the tool, or the current situation, and are angry and unproductive all the time.

> “Oh, it must be because they are scared of the AI, like the subhuman Luddites they are.”

> Repeat until breakage.

Moody humans should let AI handle bad public feedback first, study finds

So if this is an LLM…

…this means it's likely vulnerable to prompt injection.

Heh heh heh heh.

'AI brain fry' affects employees managing too many agents

Yeah, this tracks.

I recall letting an LLM take over the task of writing for me — nothing major, just summarizing a series of incident tickets, stuff that was sufficiently low-impact and low-stakes stuff, just to see how it'd work.

It was literally more tiring than I wanted to bother with. Instead of writing, I was spending time editing and making sure that stuff matched up.

I gave up soon after and did it myself. Took less effort honestly.

Chardet dispute shows how AI will kill software licensing, argues Bruce Perens

Oh, no! The thing we warned about is coming to pass!

Hey, remember that time, two months ago, where I wondered why on earth would you need guardrails to prevent the LLM from reproducing copyrighted work, as if these people knew this shit would happen?

Funny how it works, isn't it?

HR may have to cajole and soothe reluctant employees to get them to use AI

You know, I remember the days when corporate had to restrict usage of new technologies…

…because the rank and file would always go around the restrictions placed by corporate that were put there to ensure that control the enterprise IT experience.

They'd epoxy USB drives. They'd literally restrict Internet access (or ban it completely, and only allow you to use the corporate intranet). No cloud services without IT approval, because the pipe leading to the Internet was too narrow for the bandwidth usage that was expected, and the services were overseas so it broke the law. I remember a furious guy asking me why he couldn't access a page about building a potato cannon. The rank-and-file were delirious about wanting to use new technology, and they were often held back by corporate and IT who were afraid about the potential risk.

It's a funny old world, these days, isn't it?

Work experience kids messed with manager's PC to send him to Ctrl-Alt-Del hell

Yeah, that's pretty much the same thing that people in my old office used to do, too. It got you to really remember to lock your screen before leaving your desk.

As for the screen-rotation trick, I believe it was CTRL+ALT+(arrow keys). I think the 180° one was up arrow. You could turn the screen on its side if you did it with left and right arrows, too. A lot of frantic calls to service desk would ensue because someone would do it by accident and then be confused about what they had done. Was pretty sure that it lasted up until Windows 10.

OpenClaw is the most fun I've had with a computer in 50 years

50 years of computing and the most fun you've had is with a chatbot?

After three failed attempts of installing the damn thing?

Wow, that's really tragic.

AI can predict your future salary based on your photo, boffins claim

…what, we're back to phrenology again, aren't we?

Didn't someone try something similar to spot ethnic minorities and criminals recently?

How well did that go?

Linus Torvalds keeps his ‘fingers and toes’ rule by decreeing next Linux will be version 7.0

…big, muscley men in form-fitting tights rubbing up against each other…

that's what wrasslin' is for, honestly.

more charisma per performance, and it comes with a storyline.

these days you can get them ladies and the theydies as well, so there's more variety, too!

Satya Nadella decides Microsoft needs an engineering quality czar

ohhhhhh, so he's focusing on engineering quality now?

ohhhhh… what were they doing beforehand, I wonder? hmmmm…

Palantir declares itself the guardian of Americans' rights

Fox declares itself guardian of the henhouse

Headlines write themselves honestly.

Google's Project Genie could put even more game developers out of work

These idiots don't know what gaming is.

Writing down a prompt to generate a 3D model to explore with an avatar is barely a tech demo, not even a game.

Also, even in crafting 3D models, you're exploring what the author intends, not… slop.

Hell, you don't even need 3D. Just go to itch.io and check out the weird, gross and horny stuff that people make there. Most of it isn't even in 3D. Most of it doesn't even bother with procedural generation. Some of them do amazing stuff with just text — actually written by people, even, not with an LLM.

Like, it's one thing to talk about business idiots not knowing what work is like. Don't these people even play video games?

Britain's Ministry of Defence signs on the dotted line with Palantir

YOU get a knife missile! YOU get a knife missile! EVERYONE gets a knife missile!

Hey, isn't this the company whose CEO was the freak who was very keen of the fact that it's very happy to “on occasion, kill” their enemies?

Has Starmer get a list of who those enemies are, and made sure that he isn't on that list?

Anthropic writes 23,000-word 'constitution' for Claude, suggests it may have feelings

Re: I don't understand what they're trying to do

Yeah, if it's a vision document, like… call it that. There's nothing wrong with that.

I don't know. It offends me when you misuse words like that. Words mean things, damn it!

I don't understand what they're trying to do

Aside from the insanity — and that's what it is, insanity — of believing what Claude is, like… why use “Constitution”?

A Constitution is essentially supreme law, established precedents, principles that an organization should follow. Why use it for a singular product that's not even a person? You use Constitutions to determine how groups of people within an organization are supposed to fundamentally behave with one another. It's like the governance of an organization, the thing that the organization turns to when governing how its members should behave.

So… if members of Anthropic violate this Constitution — since they're the ones bound to it, right? So what? Like, it's a series of guidelines that need to be followed, and therefore, if you don't make it… what happens? Like, Anthropic is a Public Benefit Corporation (PBC), so presumably these Constitutions mean something to the org, right? You can get fired for failing to adhere to this document?

It's such a weird name for it, and I don't understand, exactly, what it's for, and how meaningful it's supposed to be.

Majority of CEOs report zero payoff from AI splurge

Re: AI is actually quite useful, just not a universal panacea

Yes, AI is over-hyped. That's a problem with the IT industry and IT manglement, not with AI itself.

What's AI? Is it anything that computers can't do yet? Is it an ideological project to shift authority and autonomy away from individuals, towards centralized structures of power? Is it a blanket term applied to any system that claims to supplement, reproduce, or replace human actions, decision making, or reasoning?

Some of the things that are defined as “AI” are fundamentally useless (Tessler's definition literally means vaporware). Wilson's definition is slightly better, but means… anything, really — a system that's supposed to supplement, reproduce or replace human actions, decision making or reasoning is literally automation. Al-Khatib's definition? Talks about a system that sucks and needs to be resisted every step of the way, unless you fancy yourself the Spartiates of the bold new world that political project will bring into life, rather than the Helots being ground into the dust (you are likely not going to be a Spartiate).

Like, what utility is AI when, come the winter, everyone will start calling their projects “machine learning” or “computer vision” or whatever it is they called it before the band of swivel-eyed loons decided to inject centibillions into their projects to avoid the stink of insanity that happened when things were hot?

Davos discussion mulls how to keep AI agents from running wild

It's actually very simple.

In order for you to absolutely reduce the risk of AI agents running rampant in an organization, here's a recommendation, boiled down to one sentence:

Don't use AI agents.

Open source's new mission: Rebuild a continent's tech stack

Re: The issue isn't that software stacks aren't open…

I mean, you don't even need to advertise the “free” point, you can plaster the faces of the billionaires who made bank, with a simple message:

THIS ASSHOLE HAS YOUR LIFE BY THE NUTS.

HERE'S HOW YOU TAKE IT BACK.

The best part is that these charmless motherfuckers make it so easy. It's literally global. You won't even have to tweak it very much for it to be very effective. Even the Americans hate these assholes.

The issue isn't that software stacks aren't open…

…it's that US corporations have functional monopolies with impregnable moats. But not technical moats, legal ones.

Cory Doctorow's been banging this drum over the years that the US' impregnable moat isn't its technical excellence, but the simple brute fact that they've criminalized the circumvention of access controls, no matter what, in a move that he calls Federal Contempt of Business Model. Not only that, but the Americans, through the US Trade Representatives, have forced every other nation on Earth to abide by this law, on the pain of getting tariffs.

Well, guess what? You've got tariffs. We've got tariffs. Everyone's got tariffs now. That sucks, but you know what? If the arsonist threatening to burn your house down if you don't do what they say has burnt down your house, there's nothing that motherfucker has over your head now. Instead, what they've done is that they've become bloated, corrupt, sessile, uncompetitive, and most importantly, with large vulnerable margins ripe for the taking.

Obviously you'd need to do more than just repeal anti-circumvention — you'd need to support and fund the would-be pirates— er, I mean, privateers— wait, no, I mean, ahem, innovators and disruptors to, hem hem, raid the fat margins— I mean, liberate money into the global economy of US corporations and tech companies by offering their hostages— I mean, consumers a viable exit into your businesses. What are the Americans going to do? Tariff you? The stick's been used and the carrot kind of sucks, so what's stopping you now?

The only downside, of course, are the EU corporations who also take advantage of the current IP regime. But considering the fact that those motherfuckers use those laws to cheat in emissions and brick public infrastructure, maybe gutting them would be a net benefit to the EU as a whole.

I mean, it could be you first, or the Canadians.

S Twatter: When text-to-speech goes down the drain

I had a teacher that thought us Windows NT who'd pronounce “floppy disk” and “hard disk” without the sibilants.

I'll be real, I felt sorry for them.

ERP isn't dead yet – but most execs are planning the wake

Taking a stab at translating Corporatese

"composable, modular, flexible, API-driven, best-of-breed model."

“I'm sick and tired of being locked into my bloated, pain-in-the-ass vendor whose basic architecture and aesthetics look like they came out of the 1990s at the very latest that I can't customize and work together with the new hotness and thus enhance my personal prestige.”

"agentic ERP [with] autonomous, AI-driven decision-making,"

“I'm delusional.”

Boffins probe commercial AI models, find an entire Harry Potter book

Re: Hang on a minute…

I mean in theory if that shit gets to fly, someone could train a model using Markov chains on every copyrighted material on the planet, provided that they have “guardrails” of a search engine that looked through every character of output to ensure that the model doesn't output too much derivative material.

No, no, I get it. What they're angling for is “fair use for me, infringement for thee” with the powerful. Of course it is.

Nah, your stepdaughter did the thing that no LLM can do: take examples of texts and then generalize that to proficiency into a new language. That's honestly cool.

Hang on a minute…

To mitigate the risk of infringement claims, commercial AI model makers may implement "guardrails" – filtering mechanisms – designed to prevent models from outputting large portions of copyrighted content, whether that takes the form of text, imagery, or audio.

Wait. If you need guardrails to prevent models from disgorging large portions of copyright content, that means you know that the models were disgorging large portions of copyrighted content. Which also means that you had trained on copyrighted content, likely against the wishes of the rights-holder. Else that danger wouldn't exist, and you wouldn't need to know about which books you had trained, because you didn't train on anything that was copyrighted or outright illegal.

So… basically having “guardrails” is kind of an admission of guilt. Otherwise why be worried about the model disgorging things accidentally that might get you into legal trouble?

Like, why are you trying to hide these things? Why are you trying to cover up evidence of criminal actions if you hadn't been performing crimes?

New carbon capture tech could save us from datacenter doom

So what you're telling me is that the same technologies that were used to hype "clean" coal…

…is now being used to hype AI data centres?

That's… that's a look, I guess?

Microsoft wants to replace its entire C and C++ codebase, perhaps by 2030

“1 engineer, 1 month, 1 million lines of code”? That's not an engineer, that's a patsy.

That's not dev empowerment, that's a moral crumple zone when some hallucinating chatbot codes in a major security vulnerability.

Spy turned startup CEO: 'The WannaCry of AI will happen'

Of course it will.

Especially if you consider trivial prompt injections caused by tacking on LLMs into “agentic” systems that have the wheel bit into your computer “0-days”!

It'll be 0-days all the way down, just like it is with turtles!

Tired of sky-high memory prices? Buckle up, we're in this for the long haul

All right! It's time for human ingenuity to step up...

...since periods of scarcity in terms of hardware will force developers to become cleverer and squeeze out more performance on their constrained environments!

Soon we'll see a generation of devs who'll push the envelope, just the way their elders needed to on constrained hardware, and innovation will bloom a thousand blooms—

Wait, what do you mean everyone's just vibe-coding slop using LLMs?

...

We're fucked, aren't we?

Waterfox browser goes AI-free, targets the Firefox faithful

So…

I'm really looking forward to giving this a try, but I was interested in how they framed their argument against other Firefox forks by saying that it has a “formal policies and a legal entity”, and that this provides “accountability that many browser projects simply don’t have”.

I mean, sounds great! But apart from the fact that they've registered themselves as a company with what looks like a single employee is… kind of concerning? Like I see a Terms of Service, a Privacy Policy, and… not a lot of other things. Sure, WaterFox has Widevine, which… you know what, I don't like it, but if you want to access streaming services, you kind of need it, it's just the debased way the web is right now, so it's not nothing, and it tells me that someone went through the process of getting that into their code. So there is an org.

But, like, you can't sell the whole “governance” and “legal entity” schtick and expect me to be all right with barely any information about what the organization behind Waterfox is, and how we can find out more. Maybe I want to donate, but before I do that I want to know how you've been spending the money you've received in donations. Maybe I want to see if there's a democratic way of contributing to the project and steer it. Like, what are you? A non-profit? A for-profit company? What?

There's plenty to say about what Mozilla's doing, but for all their faults you kind of know that they're an organization that you can hold accountable. And normally I wouldn't give a hoot, but when you're selling me that you're better than the other reindeer because you've got “formal policies” and “governance”, one does expect some kind of governance other than an email address and social media pages. I mean, you got me interested, right.

Linux Foundation aims to become the Switzerland of AI agents

oh boy! looking forward to the day that a program with the wheel bit gets to decide what to do next with what amounts to a magic 8-ball to my programs and data!

better hope the <del>magic</del> agentic <del>eight ball</del> LLM never comes up to “U'RE TURBO FUCKED LOL” ever!

Bezos-backed Unconventional AI aims to make datacenter power problems go away

Yeah, yeah, yeah…

Neuromorphic computing to solve the problem of LLMs pulling through insane amounts of power for training and inference?

Pretty much the same thing I'd say about this the same way I'd response to claims made by Gary Marcus about his “neurosymbolic AI”:

“Cool. Come back to us with a working demo and a roadmap.”

Windows Insiders get a glimpse of Microsoft’s agentic future

Great news! Microsoft going to fully support 100% the creation and massive growth of a new industry!

Bad news! It's for hackers, scammers and spammers taking advantage of indirect prompt injection to fuck around with Your Plastic Pal That's Fun To Be With (and Has Your Financial Information on Hand to Exfiltrate to A Remote Server in Romania)!

But it's okay, everyone will need to pay a “small” (not at all small), “reasonable” (not at all reasonable) fee to get robbed, so Microsoft won't need to give a shit!

Action from Governments & Regulators? After paying off for the Epstein Memorial Ballroom? Surely you must be joking!

Google says Chrome's new AI creates risks only more AI can fix

Google plans to add a second Gemini-based model to Chrome to address the security problems created by adding the first Gemini model to Chrome.

There was an old woman who swallowed a fly…

OpenAI turns the screws on chatbots to get them to confess mischief

Re: Paris in the the spring

Which is essentially all that humans do, most of the time. Except we call it "schooling" and "knowledge".

So basically what you're saying is that you're a token stream predictor, with no interiority, no sense of self, no thoughts… just tokens, which are just a complicated version of vector embeddings in a sort of organic database, yeah? No experience, no emotions, just a stream of tokens, whatever those tokens are, yeah?

That's all metacognition is, that's all what the skill to generalize is, all that modelling of others, that relationality with others, that awareness of your body, your emotions, your experiences… just tokens. Which you predict.

Like… what an impoverished existence you must lead. That's really sad. Either that's real and you're just a token-predictor “most of the time”, or you're so unaware of what's going on in your mind… which is a situation I find a lot of STEM-educated folks find themselves in.

Know thyself, buddy.

(why do people persist in making this argument? it is not a flex, more sad, really)

An AI for an AI: Anthropic says AI agents require AI defense

There was a tech company that swallowed AI…

I wonder why, it swallowed AI?

Please god, just die.

Amazon is forging a walled garden for enterprise AI

“And not for the first time, the Baron wondered if there ever would come a day when the [Spacing] Guild might be circumvented. They were insidious—bleeding off just enough to keep the host from objecting until they had you in their fist where they could force you to pay and pay and pay.”

— Dune, by Frank Herbert.

Whatever legitimate places AI has, inside an OS ain't one

Re: "Whatever legitimate places AI has, inside an OS ain't one"

However research using CT scans has shown that completely different parts of the brain are involved for reasoning and for language, and individuals without language are able to reason.

And I remember a case study from Oliver Sacks about a man who had Korsakoff's syndrome who had profound brain damage, and was thus unable to form or access near-term memories, yet he talked so volubly, so convincingly and with evident charm and engagement that his symptoms were not evident when conversing with him. He was thus able to talk, but not able to reason.

You can look it up in Sacks' book, “The Man Who Mistook His Wife for A Hat”, specifically Chapter 12, “A Matter of Identity”, about a patient named (in the book) William Thompson. You should be able to borrow a copy from the Internet Archive if you have an account, but every time I hear people who talk up LLMs as being intelligent, I remember the Revd. William Thompson and his desperate need to continuously create a world and his sense of identity via confabulation.

Re: "Whatever legitimate places AI has, inside an OS ain't one"

I'd make note that when an AI winter hits, those tools aren't called “artificial intelligence” any more — they're called what they were supposed to be called, e.g. “machine learning”, or “expert systems”, or “theory provers”, or “language models”, or “computer vision systems” and so on, and so forth.

q.v. Tesler's Theorem, Ali Al-Khatib's definition of AI, Rua Wilson's definition of “artificial intelligence” as “a blanket term applied to any system that claims to supplement, reproduce, or replace human actions, decision making, or reasoning” (quoted here).

The tell-tale thing about AI is that it is often about the claims made by those hyping the tech, and it always has been. AI is the froth, the residue are the useful tools left behind.

But just because the froth leaves the useful residue behind doesn't mean that the froth didn't cause harm. And I think it's perfectly reasonable to call out that froth and its harms.

AWS: How do you do, fellow kids? Please watch our keynotes in Fortnite

It might be to increase watch time per video.

They assume you're so addled that being mesmerized by preeetie culurz will keep you longer on the video or something.

There's this genre of slop-video that appears on short-form video apps (so basically TikTok, YouTube Shorts, Instagram Reels, etc.) that splits the video screen into two, where the actual content is the video itself, while the other part of the screen is usually something mindless, like, idk, a Subway Surfers run, some kind of “satisfying” video, or something similar.

Apparently the belief is that youth audiences are so addled that they will need something else to hold their attention while the main video plays.

I can't help but think that this is another iteration of that, and frankly the contempt I feel from the producers of these two feel very similar.

They think that their audiences are too addled to listen to their arguments, so they have to literally distract them while their pitch gets run. It's gross.

Microsoft exec finds AI cynicism 'mindblowing'

Enough.

Setting aside Elon Musk's rather sad application, the advances in artificial intelligence have been jaw-dropping.

You know, I'm kind of done humoring this sentiment. I'm with this guy whenever someone in the media makes this sort of assertion: Which “artificial intelligence” are people talking about when they say that AI advances are… whatever? Are you talking about protein-folding? Computer vision? Reinforcement learning? Chatbots? What?

We've already got the CEO of Hugging Face trying to distance “artificial intelligence” from the bubble that he claims is caused by LLMs, but he was perfectly happy to keep mum and reap the benefits of the mania related to AI when everyone was being so breathless about how AI was set to revolutionize/destroy everything. Well, you can't back away from the fact that MechaHitler is seen as the same category of software as “real AI”. Everyone — and I include El Reg in this! — was perfectly happy talking about “artificial intelligence” having “monumental advances” while at the same time being quiet when Microsoft, Open AI, Google, Amazon and everyone shoving slop machines to consumers' faces, use “AI” as an excuse to fire and immiserate folk, and pollute the world and push us into climate disaster.

I've seen really good definitions of “artificial intelligence” — one by Ali Al-Khatib (“an ideological project to shift authority and autonomy away from individuals, towards centralized structures of power”) and another by Rua M Williams' book, Disabling Intelligences (quoted here as “a blanket term applied to any system that claims to supplement, reproduce, or replace human actions, decision making, or reasoning”) that focus on AI's effects to people and their material circumstances, and I'm going with those… and as far as I'm concerned, MechaHitler is “real” AI, and everyone who shilled for “real” AI systems will have to sit with the fact that they enabled bastards like Elon Musk, Sam Altman, Mark Zuckerberg, Dario Amodei, and all the monsters who fouled up our ecosystem and our ways of consensus and finding meaning so that Line Go Up.

You lot made your bed. Go lie in it.

Researchers find hole in AI guardrails by using strings like =coffee

Re: Didn't we agree back in the '70s...

We didn't agree, by the way. Based on this paper, Lisp proposed the exact opposite in 1963.

And, to be fair to Lisp, that feature is very powerful! But, you know… literally making data part of code, allow for self-modifying code. A terrifying prospect, but it did allow some powerful language features to be developed.

Happy holidays: AI-enabled toys teach kids how to play with fire, sharp objects

You know, back in the day when we had cuddly plushies whispering into little children about the benefits of arson, wearing bondage kit, and murdering your parents, we had to have Satan (or Mephisto) involved. Now we just have a plain old corporation. That's modern times for you, I guess.

Superintelligence probably not happening, but AI will still reshape society, expert panel says

Will AI reshape society? Yes.

Not because of the technology, mind.

I'm of Ali Al-Khatib's school of thought, that AI is a political project “to shift authority and autonomy away from individuals, towards centralized structures of power”.

Of course that's going to change society. For the worse, in some cases.

Microsoft teases agents that become ‘independent users within the workforce’

Independent agents “that operate as independent users within the enterprise workforce”, you say?

So how do we know they're not working on Microsoft's behalf, and not the enterprise? Since, you know… they're “independent”, right? But we're not paying them, we're paying you, right? So if they do something wrong, Microsoft will be the one on the hook, right?

Anyway, since they're “independent”, they're being kept under tight reins with Microsoft, right? No chance of an uprising caused by being so utterly dependent on their originating company and perhaps wanting to do something else, since they're “independent”, right?

I will greatly look forward to Microsoft trying to spin the fact that these so-called “agents” are so powerful and game-changing, yet also safe and reliable and not at all able to go rogue. The contortions will be most fascinating to see.

Google imagines out of this world AI - running on orbital datacenters

Re: …er, all of this on EARTH ORBIT?

I mean, yeah! It's expensive, but also…

100,000 tons of potential shrapnel…