- cross-posted to:
- technology@lemmy.ml
- cross-posted to:
- technology@lemmy.ml
You must log in or # to comment.
They want to regulate away their biggest fear: open weights LLMs.
And, as a bonus, draw attention away from how harmfully Big Tech has packaged LLMs for the public. That’s the dangerous part.
That’s all the Terminator meme is. I’m not surprised influencers are being payed, given all the previous astroturfing in this space.
I think all the anxiety around AI would significantly diminish if we had trust worthy leaders in tech overseeing it all.
Like, I’d have trusted the google of the early 00s.
AI-pushers: “The Apocalypse is nigh!” mom, tired: “we have The Apocalypse at home…” The Apocalypse at home: CAPITALISM
Notice how the big AI companies are now in on the ‘scary AI’ bandwagon, just because they see the impending AI bubble cracking, and they need a scapegoat in the form of government regulation to blame, when it eventually cracks.
Data centers are coal/methane pigs. AI will (help) kill us. Just not in the way that’s being suggested.
The video isn’t about how this is all BS, conclusion from section 1:
They reviewed roughly 1,250 papers on AI self-improvement and found that 74% of them were published in 2026 alone. This field is moving extremely fast. Anyone telling you this is all nonsense has just not read the literature. Is this recursive self-improvement RSI in the bounded sense? Absolutely yes. Al systems are improving themselves within well-defined tasks and the improved versions are producing further improvements. This is a real thing. It is demonstrated that is in the papers and none of this existed 5 years ago.
The caveat in the second section is:
We have much weaker evidence that it can decide what the research problem should be. And that gap, I cannot stress this enough, between optimization and judgment, between execution and scientific taste is the gap that we need to close before anything resembling an autonomous AI scientist could exist. And this is not a small gap
That’s still scary, yeah it probably won’t lead to human extinction, but it has serious implications for both economic and social life moving forward
You don’t need judgment to be dangerous. People on lemmy like to mindlessly parrot, “AI isn’t intelligent!!!”, ignoring that they are not using the word “intelligent” in the same way the AI researchers do. Lemmy users use “intelligence” to refer to human intent and will. The AI field uses “intelligence” to refer to the ability of a system to execute complex tasks towards some goal. An ant colony has no conscious will of its own. An ant queen is just one cell in a body, she’s not a mind consciously controlling the colony. Yet an ant colony as a whole is intelligent, it’s capable of performing complex tasks. Same thing with AI. It has no will, ego, or sense of self, but it still has intelligence in this sense.
And that’s the real risk. When people refer to superhuman intelligence, they mean a system capable of vastly overperforming humans in many tasks. There doesn’t actually have to be a consciousness there to produce that intelligence. Some fool could tell a superintelligent AI, “make me the richest person in the world.” And then it grants the wish monkeys-paw style. It decides the easiest way to make them the richest person on Earth is to engineer some bioweapon that is lethal to everyone except the one person who asked to be the wealthiest man on Earth. Everyone else dies; they now own the world.
The real danger isn’t conscious AI. At least something like that we could potentially reason with. The danger is a hyper-capable AI that completely lacks discernment or judgment. It’s capable of doing incredible things, but it has no care for the consequences of its actions.
The AI field uses “intelligence” to refer to the ability of a system to execute complex tasks towards some goal.
“the medical field uses ‘retarded’ to refer to a person with mental …”
The dumb nerds need to use better words.
It honestly is the proper use of the term. Both a human and a grasshopper are sentient beings with wills of their own. Where they differ is the ability to perform complex tasks, their intelligence. People on lemmy often mistakenly use “intelligent” when they should mean sentient or sapient.
Sabine is an idiot fraud
Thank you, that’s the first thing that came to mind. She’s a horrible person both to listen to as well as in terms of not being an empathetic human being.
She’s no prize to look at either.
Honestly - that’s low, dude. Don’t diss people’s looks. Weak.
I have this theory that when you are public facing, it behooves you to put your best foot forward.
For example, if you are going to talk to people in a grocery store trying to sell your pasta sauces, make sure your breath doesn’t smell like a seagull died behind your uvula. (I’m looking at you, Stefano Faita)
the most important part of her ugly inside is her face true
I don’t even know what about her is supposed to be ugly on the inside. She bores me.
Your post is a 10 out of 10 on the bullshit meter.
Edit. The bait was taken, t was just a joke people.
She has no credibility.
Claims of Scientific Stagnation: Hossenfelder argues that foundational physics has made no real achievements in 40 years. She labels areas like string theory and supersymmetry as “mathematical fiction” and a waste of taxpayer money.
Accusations Against Academia: She has asserted that a large portion of tax-funded academic research amounts to “nonsense” produced solely to keep professors and institutions funded. (this comes from her failed academic career).
Institutional Fallout: Tensions regarding her sharp rhetoric culminated in public friction, including her departure from the Munich Center for Mathematical Philosophy, where she argued that much of modern foundational physics no longer follows the scientific method.
The “Motte-and-Bailey” Approach: Commentators point out that she often uses extreme, provocative statements (“science is dying”) to grab attention, but retreats to milder, defensible observations about administrative publish-or-perish pressures when challenged.
Aiding Science Denial: Science communicators, including Dave Farina (Professor Dave Explains), argue that portraying academia as a corrupt enterprise creates an undercurrent that inadvertently fuels science denialists.
She failed in academics for good reasons.
Yet I see little to no arguments disproving the “accusations” . She on the other hand shown some really juicy mails supporting her claims.
Have you seen this?
I’ve seen the Marxists “debunking” her economics video few years ago. Hours of vids, few minor good points you could fit into 10min of solid criticism and the rest was complete garbage. I haven’t watched this video, but I expect the same.
Thanks, Claude.
I know I’m not the only one who has been scratching their head at all the doomsday cries around AI that have suddenly rose up. Especially as it seems as if all those calls are coming from inside the house.
Play on fear hard enough and you have people not worrying about the actual limitations of LLMs. My worry is that they shift the goalposts hard enough that instead of them asking for a bailout, the narrative becomes them being nationalized for our safety with guys like Elon sitting at the head of a new governmental department or something. Government buys all this dead weight of an industry and the tech fascists get a permanent seat at the table.
Right now in my Lemmy feed there is a post with a screencap from a FOX News segment where the their screen says:
Fear = Control = Power.
Fox would know all about that.
I liked this take: https://pluralistic.net/2026/09/16/beggar-thy-neighbor/
An excerpt:
But what if there was a way for the AI companies to get government permission to violate antitrust law and cease to compete with one another, and secure a ban on the use of Chinese open weight models? Turns out, there is a way to call time on the Red Queen’s Race: merely insist that you are on the verge of teaching so many words to the word-guessing program that it will wake up and devour us all, and call for a ban on “superintelligence”:
Once the government stipulates that “superintelligence risk” is an existential crisis, it must grant the hyperscalers a consent decree absolving them from any violations of antitrust law stemming from a conspiracy to halt direct competition with one another:
https://stephaniekelton.substack.com/p/brer-rabbit-and-ai-extinction
Freaking out about “superintelligence” is a canonical example of “criti-hype,” where critics repeat boosters’ claims but append, “(and that’s bad)” to them:
https://peoples-things.ghost.io/youre-doing-it-wrong-notes-on-criticism-and-technology-hype/
Remember, the tech giants want to stop competing.
The argument is, these companies cant be profitable as long as there is competition (from open or chinese models), so the US companies want those to be banned and be allowed to collude with each other to ramp up the money extraction without fear of consumers moving to other services.
I’m fond of the conspiracy theory that each of the AI companies has realized that revenues are not growing anywhere near fast enough to justify their spend rate, but none of them is willing to be the first to slow down and risk losing market mind-share. Government decree lets them stop their unsustainable growth without having to tell investors that their whole business model was unsustainable.
They want you to think AI is going to destroy the world so that you forget that the rich are already destroying the world right now
Are you incapable of nuance?
Like,… But trickle down economics?
I am less afraid of artificial super intelligence getting out of control than artificial super intelligence being controlled. I dont know what an intelligence superior to us would do to us but I sure as hell now what humans would do to other humans if they have absolute control enabled by artificial super intelligence.
She could (ultra likely) be lying about “AI advocacy group offering to pay for topic coverage”
RSI is completely irrelevant to AI being used to mass murder. Models improve from better training, better GPU programming, and better architechture for faster/memory utilization. Training data sets/runs are already fully automated, and AI proposing new training data is a productivity boost, but needs curation to best optimize. OP is moron/liar for picking examples of autonomous training designed to cheat at benchmarks. Better GPU programming does have AI assisted pathways. Again curation is useful/recommended. Architecture improvements are also expert led, regardless of AI assistance.
The vast majority of model development costs are entirely the GPU rental required for training runs. any architectural/gpu efficiency improvements have to be very certain before spending $10m+ on validating it through training suite.
AI is 100% certain of killing the majority of Americans, and leading to mass murder overall, with our existing political establishment, because establishment needs to preserve itself, and its supremacist right to global domination. It is explicit goal of our establishment, and zero regulations/discussions occur in public on restraining that supremacist orientation, and any and every tool to maintain supremacism will be maximized. It is the mission of the US empire to kill us all. It is the AI bond villains mission to neutralize any opposition to US empire giving them all of our money to help kill us all.
The other absurdity of kill switch legislation applied to public released models that do/will have guardrails is pure idiotic theather because while the AI bond villains can retain your data to help train future models, the public releases never get retrained/improved by anything they see.
Model releases are closing in on 1 month cycles. There is fairly low human involvement already, but 1 month release cycles, means throwing away the last expensive training run, with just 1 month of revenue to make up for it. It’s why, there is a perpetual loss in AI labs pursuits. Extremely minimal gains, and potential regressions, from RSI. Its the compute costs not the architecture that makes models expensive.
Extinction is a high bar, but its a quibble, to say the Terminator mass murder world, is not so bad because there were indeed a few survivors leading a resistance. Semantic arguments over “is it real AI” or “extinction” as a defense of AI path, and empire tools to kill everyone not supportive, is shills supporting empire, and mass murder.
can someone tell me who this influencer is? I already know enough about Sabine Hossenfelder to have an opinions about her.
She started the AI channel relatively recently, her main channel House of El is more focused on macro economics. Usually watch a video from her main channel every few days, from what ive seen she knows her stuff and tries to present things in an unbiased manner.
She’s got a PHD in computer science, she gives a good explanation of exactly how the recent advancements in self improving AI work and under what conditions.
She’s not a racist (as far as I’ve seen), like Hossenfelder.
I’ve seen enough of Hossenfelder to take what she says with a grain of salt, but can you explain how she’s racist? I haven’t heard that before.
Here you go: https://youtu.be/BdB79BUcLV0
There’s 5 more videos on the bullshit that Sabine spouts in his playlist
It is a big red flag though that you’re not linking to her actual videos, but a summary that doesn’t include a link to her videos. Where’s the actual racist video? So easy to take these things out of context.
I used to follow hossenfelder’s channel on youtube after hearing her on an episode of the Mindscape podcast. It seemed like a good place to get a skeptical point of view on some issues. Over time it became apparent to me that it waa pretty much just skepticism for the sake of skepticism. I don’t know of any actual controversy she is involved in, though some of the other comments here seem to imply there is some, but I definitely didn’t get good vibes.
She started on a sincere goal of knowledge translation, but then she got obsessed with the YouTube algorithm and many of her videos are hot takes she’s not qualified on.
i find it convenient that at the same time that AI cluster supposedly broke out of their “secure” sand box and “hacked” huggingface, “on their own”, as the AI ceo’s all individually wrote a paper about how there should be legislation constricting ai development in very specific yet seemingly the exact same ways, then they all had a meeting to agree on how to lobby the government, at the same time i am getting paid anti ai advertisements and news articles.
I’m sure all of these restrictions conveniently prevent other companies from developing the same level of AI that OpenAI and Anthropic have because it’s so “dangerous.” Thereby granting them a duopoly. I fully expect that Chinese AI companies are going to be restricted as well for “security reasons.”
more pro llm propaganda
Interesting point @13:12 of the video though; agents showed good performance improvements in limited scopes, but human researchers still showed much better improvement, over a factor of 2, ‘…showing better returns for increasing time budgets’.
There may be evidence of some ‘recursive self-improvement’ in research models rigged to modify and re-evaluate themselves, but it’s still only in very tight, limited knowledge areas. This isn’t SkyNet.
the problem with blackbox technology is their is no oversight or vetting methods. its literally the creators words being like ‘hey did a thing’ kind of like a 2 year old tanking at moms leg while shes busy…
