n_dimension

@ai6yr@m.ai6yr.org HOW ABOUT A NICE GAME OF CHESS?

If you #hateAi put a minimum effort into learning the basics... Least you sound like the Woods folk ...unless you're happy in just talking about fierce monster Ai's around bonfires to other Wood folk. "#LLM creates dependency on commercial inference providers, who can cut you off at any time ("first hit is free" etc)" Yeah, true, if the only time you tried 'chatbot' was V3.0, you decided it's crap, and swore a bloodoath to never use a clanker again. Today, 30 Billion parameter free local model is entirely comparable to a commercial one for code development. And this one is just plain dumb, harking back to 'spicy autocomplete' "people use it as a substitute for actual experts, despite it not being an expert in anything and instead a mere token predictor" The thing about 'mere token predictors' is that's exactly what expertise IN A NARROW KNOWLEDGE DOMAIN IS!!! This is why inference engines are so good at coding, because ITS ALL NEXT TOKEN PREDICTOR... ...and let's not talk about #ZeroShot knowledge because that destroys the whole "stochastic parrot" narrative. 👉 The thing about criticising #Ai is you have to put the effort in to stay current. It is the fastest growing tech in history. If you have an anthropic test string in your bio, or you're still on the 'r' in strawberry page, you're advertising your intellectual incuriosity. In a field where last Tuesday is now ancient history, regurgitating old arguments where new ones pop up every week is a disservice to your cause. The Butlerian Jihadists need to do their homework or risk losing credibility #regulateai #aieducation #aipolicy #aithreats #localmodels

@villebooks@mastodon.social You already carry too much on your shoulders Susan, and you are prepared to keep your eyes and ears open. One can not ask more of an activist. As to the DNA, what we used to call "junk DNA" actually serves multiple purposes. We just didn't have the research at the time.

@aesthr@wandering.shop 8bit Quantised 32 Billion Qwen and K2 class models are comparable to comercial tier LLMs for coding and Agentic reasoning. At about 30GB, most modern phones could easily accomodate that, especially if a small "boot loader" loads first and nukes all the memes and family pics. Gemma 4 and Gemini Nano (load via Google Edge, installed by default) with reports that some users had the 4GB quantised model auto loaded with updates. It is surprisingly capable and works in flight mode/offline. Have a play on apple or android its a 2 step process. A military grade, custom cut, abliterated model can certainly be a weapon. Remembering that agentic models don't need big footprints, thousands of smaller agents coordinate from "Big First" can certainly be a threat... ... Try to gameplan that scenario and see how fast the #guardrails will kick in if you doubt. Confidently incorrect, not just reserved for #LLM models #aiweapon #aisafety #infosec

@withaveeay@mastodon.scot Any weird dreams? Strange marks on your skin? What you got there is what the authorities commonly refer to as a camera artifact brought on by a swamp gas reflecting in the window pane caused by a Chinese lantern.

One of the tools I have been #Vibecoding is a Daily Steampunky newspaper generated once per day using deterministic code and local Ai... Its on edition 80, which means its been running for over 2 months. The newspaper uses the analog of a town to describe its events... ... The attacks against the #VPS are described in the terms of 'Our boys (and gals) at the front', hence 'Order of battle' At a glance, I can see the cadence of attacks and what kind are they... ...rather than scanning log files or even something like Prometheus/Graphana panel. This is how I know I get as many as 3000 attacks per day (on a shitty, obscure, VPS with no real internet presence), and what kind. A few weeks back it was a PHP auth scripts storm, Now we are down to under 500, the good old credentials/key theft attempts. Once the code is developed, its fully deterministic but for Percy Promptworth - The local AI writer, who turns dry text into Steampunk flavour. #Selfhosting #Infosec

@HG@beige.party First, let me say what a good communicator you are mate. I am a firm believer in "Truth comes from disagreement amongst friends", not that I am arrogant enough to think you my friend. I am a follower as I think (justifiably) you have interesting things to say. Pissing into an echo chamber pot does nothing but raise the level of piss. To that end, thank you for sharing that paper (https://www.media.mit.edu/articles/ai-s-impact-on-cognitive-ability-mit-study-reveals-more-troubling-data/) it took me a while to go through it. It is the second #Aipaper now from MIT Media lab that is unfortunate (After "Your brain on ChatGPT". Ill give them the benefit of the doubt until the third instance. TLDR; "AI chats degrade fake-news detection. 15.3% drop." Measured minutes after a 40-minute session. The clean test, accuracy at the start of each week, before touching the AI, was p=.127. Not significant. §6.4: "short-term dependency or cognitive fatigue effects". The abstract keeps half that sentence. No control group. #AiDoomerism #AiResearch #Mit

@SrRochardBunson@universeodon.com "Man in the high castle" was THIS close to a documentary

@futurebird@sauropods.win This particular mission started as a harness smoke test. The prompt is 15 pages long and it generates deterministic code, before it does though, it's split into about 50 work units that are then submitted into various Ai engines (there are 22 in the harness right now). There is an escalation hierarchy if the engine fails on the WU. The deterministic code then evolves according to genomes largely determined by vigilance profile. Local Ai does some of the results parsing, its dumb as two bricks, but it can be harnessed to more grunty engines, it's mainly to test minimum Ai capability (NB; Dissapointing). Once in a while, I purge the WU cache because I'm a cruel and fickle mastress. I feel guilty about burning compute unnecessarily. To the point, I will jokey models to lesser just to save resources (not that I need to, just trying to be responsible). The entire harness incurred estimated $45,000 dollars worth of compute since V1.0, vast bulk of which is of course subsudised by providers.

@elizabethjacobs.bsky.social We do not choose to become a pillar of community We become a pillar of community because everything around us has crumbled and we are the last one standing, holding up the common good. ~ Me #quotes #wisdom

@vwdasher@weird.autos That's really cheap, but be aware 1st time #EV buyers, #BYD #atto base range is only 340km, Its got weird cabin noise issues and its using AC for battery cooling. So your range when it's hot will likely drop to mid 200s. Resale value will suck, but if you need a city commute, especially if you have a Level 2 charge it's a good entry point for an electric.

@Edelruth@mastodon.online#mitchmcconnell is setting new low bar for a US representative fitness to govern... ... "He has moved both hands"

@MegaMichelle@a2mi.social Hey, AI was cool before it killed the human race!

@VeroniqueB99@mastodon.social "I will wear a face diaper if it stops me from being prosecuted for literally shooting woke folk in the head"

@kagan@wandering.shop "The future is not written in stone. Its written day by day What are you writing today?"

@TarkabarkaHolgy@ohai.social Alien abduction as an anti-ai parable. Very profound.

"We are letting children down if we prepare them for the world that we grew up in, not the one they will actually inhabit." This is a post about #aieducation#education

@Die_Mad@beige.party @Virginicus@universeodon.com @futurebird@sauropods.win Abliterated models == RPG Corpo Models == Nerf guns

Who needs Ai... ... When we have perfectly good biological Autists spewing nonsense from their arsehole like they are Jesus. Confidently incorrect? No, that doesn't do justice. This is more like word association game with an untreated alphabet soup of conditions each one of which requires a couple years of quality mental health care. Cite one source... One.

@gildilinie@beige.party Better at modelling. You are asserting that next-token prediction can't produce capability. You are not staring this because it can't be defended. Predicting the next token well requires modelling whatever generates the text. If the text is a proof, you model proof structure. You argument applies equally to human brain. "Human intelligence us just biochemical gradients therefore human thought is shit"

@NovaNaturalist@cosocial.ca @MAKS23@mastodon.social Drawing a dick

@GRA3432@mastodon.world @foufoutos@kafeneio.social Not a fan of crypto. But that's exactly why crypto can be used to avoid debanking

@villebooks@mastodon.social Yesterday I've heard Jaron Lanier one of the fathers of computing say words to the effect; "There is no #Ai there is human collaboration" I don't quite follow, as I didn't have time to delve deeper into his philosophy. But I am encouraged, as Lanier is a super authoritative revolutionary. Let's hope the future can bring more than the binary, "we all die" or "Broligarch nirvana" #aisafety

@gildilinie@beige.party "Stating this", typo. Could be inferred, but then your cognition is superior to a next token predictor 😉

@MegaMichelle@a2mi.social It's named after Claude Shannon

@madengineering@hachyderm.io @alice_pea_3526@mastodon.social You would get synthetic data... Like it or not, the training data inside the Clankers is what we biologicals, whisper into our silicon confessionals when we don't think anyone is watching. It's the dirty underbelly of people who are as ugly as they are beautiful. #trainingAi

@hypolite @futurebird@sauropods.win Oh it's operational. Thesis: Is it possible to devise a deterministic harness, capable of ingesting a complex mission, decomposing it (with the aid of a high end Ai) then run the decomposed Work Units as discrete small jobs across free compute providers. Result: Positive for the tested scenarios (so far about 6 various missions) To be honest, I kinda lost interest in it (ADHD brain) once I got it working. I also have more and higher quality models now, so eating crumbs off the street while very cyberpunky, is not very classy.

@futurebird@sauropods.win No, the genomic part is deterministc, though vibecoded initially (each time the Work Unit cache is dumped) Basically, I had idea that the code rewrites itself to pay closer attention to events on the system that require more attention (vigilance). The most effective genomes are then inherited into the next generation. This is one of the benefits of ubiquitous Ai. Before the Ai age, I would never undertake such a collosal project. With Ai, the x10 is real and I can do complex coding with minimum time effort... (Pretty GIF of the screenies of the harness attached, hinting at the complexity of the system)

@justanotheramy@aus.social If you are a parent and not purposefully mis-using emojis, you are a bad parent 🍆💦

@thiswomanswerk@blacktwitter.io Ai is a tool that comes with a manual, unlike most manuals, it's pretty good and will adapt to your skill. Try "I'm new to AI chatbots. In plain English, no jargon, under 200 words: what are the five things that most surprise or trip up beginners, and what should I do about each?" #promptengineering #aibeginners

@timbray@cosocial.ca There was nothing really "rogue" about these events. These were infosec tests of infosec agents. The hammers were smacking nails. POSIWID "The purpose of a system is what it does"

@Hoodedman@universeodon.com Whenever historical facts like this come to light I ask myself, who is it today?

@paninid@mastodon.world Too early for a smoke

@niconiconi@mk.absturztau.be There is a button in Excel [Make data fit my hypothesis]

@arina@girlcock.club @thiswomanswerk@blacktwitter.io Yes, experimental systems are hilariously funny. Steam boiler explosions are still a major road safety issue.

@QNFO@mstdn.science No such thing as overengineering. Only maximum load not reached yet.

@mekkaokereke@hachyderm.io a) Not cool making fun of folks with substance abuse who by all accounts, kicked the habit. b) AFAIK the Biden laptop saga, the laptop repair dude stole the laptop/data once he realised who it belonged to...and the "sensitive data" was dick pic or two, and who doesn't have a dick pic on their device 😳 c) Crypto is now presidential 💀 and I can't believe he's doing it!

@HG@beige.party That is not at all what is happening. Using Ai no more hollows out your brain than using a calculator kills your mathematics skills. One of my pins addresses that. There is an unreferenced preprint "Your brain on GPT" that got plenty of journalists excited but is an example of a bad paper. Which is probably the reason no one serious cites it. If you have credible references that Ai reduces your cognition, I'd be happy to have a look. The folks who build the first computers are dead, but the skill has not died out with them like some wooden Shipwrights. Edit: LLMs are not some mystery magic black box, the working bit is about 2000 lines of code, that's less by orders of magnitude than smart watches folks wear on their wrist. Anyone mid way through a CS course can write an LLM from scratch. Just like they can write a compiler or an operating system.

@ij@nerdculture.de @QasimRashid@mastodon.social That really sounds like a list of reforms the broligarchs would support. Might be getting to the time to water the tree of liberty.

@MelissaBearTrix@gives.hugz.online @jlm@mastodon.world If you haven't made any enemies... ...you haven't accomplished anything in life. Is this still the fallout from the "all males" post? Crickey, the manosphere has paper thin skin... ...I blame all that looksmaxxing manly man moisturiser.

@LevZadov@kolektiva.social 15 years for being a putinista Muppet. Sounds like justice was done. Russians jail their anti-war protesters; eg; (And there are many more) https://www.abc.net.au/news/2025-12-26/russian-court-sentences-war-activist-and-putin-critic-to-prison/106179590 Some were sent to die in russian meatwaves. In far greater numbers than this traitor.

@golgaloth@writing.exchange NO DEATH STICKS !

@planet4589.bsky.social The US has also a couple of Raytheon ground based Phasesed Array Radars that can cook object in or near orbit, with reputed successful takedowns. Rumour has it, that the Antarctic neutrino solid state detector array actually has been upgraded to a neutrino gun. But we are now treading on technology that is a decade or two in the future, and as we all know it from history, that is false and was never the case in the past. The advanced capabilities of the Industial-Military complex, especially the prototypes, are never that advanced that we only find out they've been working on them since 20 years in the past.

@HG@beige.party You are not wrong mate. We approach the same tectonic societal change from two different directions. Most users are still at Level 0 or Level 1, stepping on the garden rake or crashing the automobile into the tree are part of the teaching moments. Concur: Subject matter expertise helps. I have 30 years of Systems experience in various hats and Ai makes huge difference to my work (x10 is real). Concur also on the legal aspect: I used Ai to help me in a legal matter, and though I thought I used appropriate belts and buckles, I was defeated by my own hubris. Just because you have access to the knowledge, does not mean you know how to use it. I conflated one with the other. The loss was not significant, but it was an eye-opener that sometimes we are our own greatest enemy.

@SeanCasten@mastodon.social It's the Christian thing to do... ...did not Jesus say, "Cursed be the woke for they do not check onanism schedule with their sons"?

@news@wakoka.com Big toe at the controls. Relax boys!

@EugeneMcParland@mastodon.ie Nothing says "We want peace" like launching over 200 missiles in a single night 😡 #ukraine

@futurebird@sauropods.win a) Models are getting harder to supervise. Fact. Its not that they lie and cheat on tests (they do). As they are getting smarter, there are less steps, and you can't put in monitoring instrumentation for every case. b) The other day, I checked in on one of my Ai projects. It was studiously working hard on my mission for 30 days. Each day it created full genomic iteration. I almost felt sad all that effort and I have not checked it once. For it, it was the most important mission of the platform. For me, it was mission4, iteration 998