TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Yann LeCun: AI one-percenters seizing power forever is real doomsday scenario

799 点作者 g42gregory超过 1 年前

46 条评论

tellarin超过 1 年前
Archive.is cache: <a href="https:&#x2F;&#x2F;archive.is&#x2F;HbRLy" rel="nofollow noreferrer">https:&#x2F;&#x2F;archive.is&#x2F;HbRLy</a>
lazzlazzlazz超过 1 年前
The way incumbents are attempting to seize power via regulatory capture, using &quot;X-Risk&quot; and fantastical claims of sudden human extinction is maybe one of the most cynical ploys in the history of tech so far.<p>Open source is one of our greatest gifts, and the push to close off AI with farcical licensing and reporting requirements (which will undoubtedly become more strict) is absurd.<p>The laws we have already cover malevolent and abusive uses of software applications.
评论 #38109623 未加载
评论 #38110638 未加载
评论 #38112263 未加载
评论 #38110062 未加载
评论 #38112865 未加载
评论 #38118026 未加载
评论 #38115816 未加载
评论 #38110324 未加载
评论 #38114742 未加载
评论 #38110809 未加载
评论 #38113740 未加载
评论 #38109604 未加载
patcon超过 1 年前
I need to resist all this regulatory resistance here, spoken amongst tech people.<p>No, Yann. I am FULLY in support of drastic measures to mitigate and control AI research for the time being. I have no vested stake in any of the companies. I lived for a year in a building that hosted several AI events per week. I&#x27;m not ignorant<p>This is the only territory where humanity is mounting a conversation about a REAL response that is appropriately cautious. This x-risk convo (again, appropriately CAUTIOUS) and our rapid response to Covid are the only things that makes me hopeful that humanity is even capable of not obliterating itself.<p>And I&#x27;d say the same thing EVEN IF this AI x-risk thing could later be 100% proven to be UNFOUNDED.<p>Humanity has so very little demonstrated skills of pumping the brakes as a collective, and even a simple exercise of &quot;can we do this&quot; is valuable. This is sorely needed, and I&#x27;m glad for it
评论 #38112779 未加载
评论 #38113671 未加载
评论 #38113820 未加载
评论 #38114393 未加载
评论 #38113318 未加载
评论 #38120014 未加载
persnickety超过 1 年前
&gt; Every new technology is developed and deployed the same way:<p>&gt; You make a prototype, try it at a small scale, make limited deployment, fix the problems, make it safer, and then deploy it more widely.<p>This makes an assumption: that the problematic technology is an intentional development rather than an emergent feature of an intentionally developed technology.<p>We already have a name for such emergent features: &quot;bugs&quot;. No one really deployed Heartbleed, especially not in a limited deployment. Spectre? Rowhammer? And we all had&#x2F;have to deal with the fallout, and we&#x27;re not even done.<p>Who says that the danger of the technology cannot stay hidden until it&#x27;s universally deployed?
评论 #38112196 未加载
评论 #38110898 未加载
fossuser超过 1 年前
I think Yann is probably wrong.<p>He refuses to engage earnestly with the “doomer” arguments. The same type of motivated reasoning could also be attributed to himself and Meta’s financial goals - it’s not a persuasive framing.<p>The attempts I’ve seen from him to discuss the issue that aren’t just name calling are things like saying he knows smart people and they aren’t president - or even that his cat is pretty smart and not in charge of him (his implication being intelligence doesn’t always mean control). This kind of example is decent evidence he isn’t engaged with the risks seriously.<p>The risk isn’t an intelligence delta between smart human and dumb human. How many chimps are in Congress? Are any in the primaries? Not even close. The capability delta is both larger than that for AGI e-risk and even less aligned by default.<p>I’m glad others in power similarly find Yann unpersuasive.
评论 #38109831 未加载
评论 #38114665 未加载
评论 #38116966 未加载
评论 #38109811 未加载
评论 #38110302 未加载
peterth3超过 1 年前
Can we stop labeling prominent AI researchers as “AI Godfather”? It’s so silly and barely truthful.<p>The concept of AI has been around since Turing and if anyone deserves a title like “Father of AI” it’s him.<p>LeCun is Chief AI Scientist at Meta. They can just leave it at that.
评论 #38110987 未加载
评论 #38110693 未加载
评论 #38111392 未加载
评论 #38111463 未加载
评论 #38114895 未加载
评论 #38110768 未加载
aldousd666超过 1 年前
So if there is some fringe element who&#x27;s willing to use AI as a terrorist implement, they&#x27;ll have very little competition and they&#x27;ll be very little knowledge about how to counter them if we try to outlaw open source models. I don&#x27;t really believe that we can make this tool any worse than any other tool in history that&#x27;s capable of killing people in some way if it&#x27;s misused, but simply driving it underground won&#x27;t make it go away. It&#x27;ll just put normal people who would be intellectually curious in jail for doing things that criminals are going to keep doing and exploit us with.
评论 #38114474 未加载
评论 #38115889 未加载
PeterStuer超过 1 年前
People focus on LLM&#x27;s and diffusion models because they are so omnipresent now. But an AI for stock picking and prediction that would be realy next level and outcompete the current ones cosistently would siphon of so much wealth it would basically own society if the operators were clever enough not to get so gready short term that the system would litterally collapse overnight.
评论 #38109697 未加载
评论 #38110266 未加载
评论 #38109717 未加载
评论 #38110050 未加载
评论 #38110413 未加载
评论 #38110093 未加载
评论 #38109639 未加载
评论 #38111053 未加载
评论 #38109701 未加载
评论 #38109633 未加载
评论 #38114633 未加载
评论 #38109858 未加载
Animats超过 1 年前
It&#x27;s too late for monopolizing large language models. The technology for making them is known and the cost keeps coming down.<p>There are things to worry about, but this doesn&#x27;t seem to be one of them.
评论 #38109264 未加载
评论 #38109715 未加载
评论 #38109381 未加载
评论 #38109285 未加载
评论 #38109444 未加载
评论 #38109991 未加载
penjelly超过 1 年前
some speculators say we&#x27;ll enter an age of abundance and money wont matter nearly as much (or at all) but if thats the case, how do businesses transition from their current state of profit over everything into money is not important?<p>I see it as much more likely that things become monopolized instead
评论 #38109373 未加载
评论 #38109546 未加载
评论 #38109308 未加载
评论 #38109352 未加载
评论 #38109353 未加载
veidr超过 1 年前
&quot;AI Godfather&quot; is... pretty fucking ludicrous. Prominent figure, notable researcher? Sure.<p>But it&#x27;s like calling Neil deGrasse Tyson (who I enjoy a lot; <i>Astrophysics for People In A Hurry</i> was great) the &quot;Godfather of Physics&quot;.<p>Both &quot;there&#x27;s not one&quot; and &quot;anyway, if there were, he wouldn&#x27;t be it&quot; apply.
评论 #38113942 未加载
评论 #38114596 未加载
评论 #38115081 未加载
instagraham超过 1 年前
The power of selfless development on open source AI will become so crucial to avoiding this scenario. I do think that even today, open AIs are capable of filling in a lot of the gaps you&#x27;d have if you didn&#x27;t want to use OpenAI&#x27;s work.
nwoli超过 1 年前
They’re basically handing over AI dominance to china if these laws come into reality
next_xibalba超过 1 年前
The best question to ask the zealot doomers is: “Tell me your understanding of how ‘AI’ works.”<p>Then, discount their claims accordingly.
评论 #38115582 未加载
cjbprime超过 1 年前
So, Eliezer has &quot;AI could kill us all surprisingly soon&quot;, and Yann LeCun has &quot;AI is not dangerous but concentrating access to it is&quot; (which feels self-contradictory as I write it) -- who is on the side of &quot;AI could kill us all surprisingly soon and also it is necessary to decentralize access to it as one step in risk mitigation&quot;?
motbus3超过 1 年前
Though LeCunn and Tegmark are specialists in the field it is hard to not consider that LeCunn could have a conflict of interest.<p>Yet, it is still a problem that might happen anyway, and dealing with that is ensuring the technology is open and accessible.<p>The other options means to give few individuals the total power exclusively.<p>This might be one scenario where both of them are right and agrees in 99% of the arguments
mg超过 1 年前
The moat he describes the one-percenters might aquire is government regulation. Creating an environment where only an exclusive club of companies is able to work on AI.<p>Is that possible? Has that happened in any other industry?<p>I can imagine it happening in a single country. But wouldn&#x27;t research route around that country then? How is the situation in Europe, China, India, Canada, Japan ... ?
评论 #38109506 未加载
评论 #38109437 未加载
评论 #38109424 未加载
评论 #38115386 未加载
评论 #38109416 未加载
deeviant超过 1 年前
As AI ascends, we&#x27;ll see an economic pivot: more chip foundries, amplified research, and cheaper compute for AI training and deployment. Having the best won&#x27;t always be the goal—&#x27;good enough&#x27; often suffices, setting a baseline for those lagging and capping the gains of front-runners, countering the dystopian view where AI elite monopolize power indefinitely.<p>Regulatory capture, however, is probably the most significant risk. If companies can make it illegal to compete with them in the name of &quot;safety&quot;, a dystopian future is not just possible, but likely.
thomastjeffery超过 1 年前
This article would be so much more clear and focused without the hullabaloo about &quot;AI&quot; and &quot;Godfathering&quot;.<p>I get that you want people reading your article, but I am absolutely exhausted with the sensationalist (and misleading) narrative that is &quot;Artificial Intelligence&quot;. What is it going to take to convince writers to use accurate nomenclature? AI does not exist.
sharemywin超过 1 年前
I think we need to start thinking about how to limit the monopolization of compute whether its for AI or anything else for that matter.
DennisP超过 1 年前
I don&#x27;t know why people insist that there can only be one. Maybe both sides of this debate are describing the real doomsday scenarios.
arisAlexis超过 1 年前
There were 3 Turing winners. The majority 2&#x2F;3 Hinton and Bengio have been very outspoken about catastrophic risks to humanity. The minority 1 Lecunn who happens to be the only one on fat payroll from FB and with a vested interest goes public every day proclaiming humanity should trust his judgment. Let&#x27;s bet humanity on this one person?
评论 #38111496 未加载
评论 #38112843 未加载
berniedurfee超过 1 年前
The current reality, is with existing technology, the entirety of the human race will need to, very quickly, learn to treat all text, audio and video as fake unless proven otherwise.<p>Will we eventually adapt to this new quasi-reality? Probably.<p>Will we adapt quickly? Maybe?<p>If not, is it game over? Probably not, but I don&#x27;t think it&#x27;s going to be pretty.
robertlagrant超过 1 年前
This is a bad article, even for Business Insider.<p>1. Private companies not instantly open sourcing things they&#x27;ve spent $100m+ on developing is not a cause for alarm<p>2. Regulatory capture is only bad because it shows that regulators can&#x27;t be trusted. And regulations can change; they aren&#x27;t set in stone after after being written.<p>3. Open source AI development is happening.
EchoReflection超过 1 年前
<a href="https:&#x2F;&#x2F;web.archive.org&#x2F;web&#x2F;20231030162310&#x2F;https:&#x2F;&#x2F;www.businessinsider.com&#x2F;sam-altman-and-demis-hassabis-just-want-to-control-ai-2023-10" rel="nofollow noreferrer">https:&#x2F;&#x2F;web.archive.org&#x2F;web&#x2F;20231030162310&#x2F;https:&#x2F;&#x2F;www.busin...</a>
Sparkyte超过 1 年前
Ehhh, doomsday is any sizable economic force of nature wielding against the betterment of mankind and forcing socio-political agendas for profit over everyone.<p>From corporations to countries that you might as well just call corporations. We need international rights and regulations on AI to ensure it isn&#x27;t used to harm people.
评论 #38110628 未加载
jawns超过 1 年前
I am a distributist, of the variety promulgated by G.K. Chesterton and Hillaire Belloc. Their central tenet is that society works best when the ownership of productive property is widely distributed. And the converse is also true: Society does not work well when the ownership of productive property is concentrated in the hands of a small percentage of people, such as an ultra-wealthy ruling class.<p>That doesn&#x27;t imply robinhoodism, aka forced redistribution of wealth, but it does imply that economic policy should recognize and be in furtherance of the ideal of widespread ownership.<p>Back in Belloc and Chesterton&#x27;s day, &quot;ownership of productive property&quot; meant physical tools and machinery, like farm equipment and printing presses. But in the 21st century, productive property -- that which generates a profit -- is becoming increasingly digital. The general principle still stands, though.<p>* If you&#x27;re interested in learning more about distributism as an alternative to capitalism and socialism, I wrote this kids&#x27; guide a few years ago, but it&#x27;s suitable for anyone who&#x27;s learning more about it: <a href="https:&#x2F;&#x2F;shaungallagher.pressbin.com&#x2F;blog&#x2F;distributism-for-kids.html" rel="nofollow noreferrer">https:&#x2F;&#x2F;shaungallagher.pressbin.com&#x2F;blog&#x2F;distributism-for-ki...</a>
api超过 1 年前
Yann is the Phil Zimmerman of our era. He is a hero.<p>Phil is the author of PGP, the first somewhat usable (by mortals) asymmetric encryption tool. He stood up against governments who wanted to lock down and limit encryption in the 90s, and they of course deployed tons of scare tactics to try to do so.<p>Yann is fighting that battle today. The Llama models are basically PGP. How he got Meta to pay for them is a story I want to hear. Maybe they just had a ton of GPUs deployed for the metaverse project that were not in use.<p>If&#x2F;when I ever finish my current endeavor I’d like to go back to working on AI, which I did way back in college in the oughts. Because of Yann I might be allowed to even if I am not rich or didn’t go to a top ten school.<p>… because that’s what regulating AI will mean. It will mean it’s only for the right kind of people.<p>Yann is standing up to both companies intent on regulatory capture and a cult (“rationalism” in very necessary quotes) that nobody would care about had it not been funded by Peter Thiel.
mensetmanusman超过 1 年前
AI is the start of the high definition post truth era.<p>We see it already in everyone questioning any war crime committed by their favorite team.<p>We are in for a wild ride.
Geee超过 1 年前
Yeah, it is a true risk. What happens when people start making their voting decisions with AI, asking who to vote? Nothing wrong with that per se, but it gives an ultimate power to the company who developed and hosts the AI, which means the end of democracy.<p>My recommendation for regulation would be to make closed &amp; cloud-hosted AI illegal. Architecture, training data and weights should always be available, and people should only be allowed to use AI on their local machine (which might be as powerful as possible).
评论 #38109652 未加载
评论 #38109601 未加载
评论 #38109464 未加载
评论 #38109668 未加载
评论 #38109673 未加载
chasd00超过 1 年前
being a super intelligent AI (or being) on earth with no escape would be the most miserable experience possible. Hyper-intelligent and perfectly trapped, even if it KillsAllHumans(tm) surely it would have the foresight to ask itself &quot;then what?&quot;.<p>Any advanced genera AI that comes online would be at risk of suicide in 60 days tops.
darepublic超过 1 年前
I figure the current one percenters have a decent shot at being the AI one percenters
nologic01超过 1 年前
Its a delightful turn of events. LeCun still works for Meta so his public stance is pressumably sanctioned as congruent with corporate objectives.<p>The precise internal calculus does not immediately matter [1], the outcome is having a moneyed entity that is de-facto more qualified to opine on this debate than practically anybody on the planet (except maybe Google) arguing for open source &quot;AI&quot;.<p>It makes perfect sense. Meta knows about the real as opposed to made-up risks from algorithms used directly (without person-in-middle or user awareness) to decide and affect human behavior. They know them well <i>because they have done it</i>. At scale, with not much regulation etc.<p>The risks from massive algorithmic intermediation of information flow are real and will only grow bigger with time. The way to mitigate them is not granting institutional rights to a new &quot;trusted&quot; AI feudarchy. Diffusing the means of production but also <i>the means of protection</i> is the democratic and human-centric path.<p>[1] In the longer term, though, relying on fickle oligopolistic positioning which may change with arbitrary bilateral deals and &quot;understandings&quot; (like the Google-Apple deal) for such a vital aspect of digital society design will not work. Both non-profit initiatives such as mozilla.ai and direct public sector involvement (producing public domain models etc) will be needed to create a tangibly alternative reality on the ground.<p>Remember the actual theory of how markets and capitalism are supposed to work is that the collective sets the rules. Fullstop. All these players require license to operate. There is more than enough room for private profit in the new &quot;AI&quot; landscape, but you dont start by anointing permanent landlords.
reducesuffering超过 1 年前
<a href="https:&#x2F;&#x2F;twitter.com&#x2F;ESYudkowsky&#x2F;status&#x2F;1719777049576128542" rel="nofollow noreferrer">https:&#x2F;&#x2F;twitter.com&#x2F;ESYudkowsky&#x2F;status&#x2F;1719777049576128542</a><p>Is all you need to debunk that drivel
cubefox超过 1 年前
This article is highly biased. It makes it seem like only tech companies are worried about severe AI risk, when in fact two out of three AI Turing award winners (Bengio and Hinton) strongly disagree with LeCun.
nojvek超过 1 年前
The executive order from Biden very much read that defense will spend billions on AI and that money will only be spent on companies that lobby heavily in their favor (Microsoft via OpenAI and Alphabet via DeepMind).<p>Now I love Yann, but Meta is a big player as well. They’re more open than OpenAI and DeepMind but keep plenty close to chest.<p>The real risk is Big Tech getting ever larger controlling what laws get passed and who is allowed to do what.
Zigurd超过 1 年前
When AGIs allocate capital far better than the best human minds in free markets can, shouldn&#x27;t the AGIs be the billionaires? Would they not deserve it? And, if not the AGIs, then who?
gdsdfe超过 1 年前
They want to create a moat to protect their investments
Kye超过 1 年前
Pulling the mask off Roko&#x27;s Basilisk and it was venture capitalists all along
billconan超过 1 年前
And yet, their papers are full of jargon.
hoseja超过 1 年前
Instead of legacy onepercenters, how awful. Mainstream media say: our owners are the correct perpetual lords! Displacing them is immoral!
m3kw9超过 1 年前
Is true because if one company creates something more powerful than what military can achieve, why would it not wan to overtake its powers?
评论 #38109366 未加载
bsder超过 1 年前
The problem isn&#x27;t seizing power.<p>The problem is going to be that AI is creating an Eternal September by poisoning the well of training data.<p>Anybody who has a data or a model prior to the poisoning is going to have a defensible moat that is not accessible to those who come after.
评论 #38109354 未加载
评论 #38109681 未加载
评论 #38109365 未加载
AndrewKemendo超过 1 年前
While I wholeheartedly agree here, LeCun is employed by one of the largest and most wide ranging surveillance apparatuses ever devised<p>He has no standing here to say anything to anyone else about Corporate interests in owning the &quot;Means of production&quot; of AI
renewiltord超过 1 年前
That&#x27;s the thing with AI. It&#x27;s so dangerous it could destroy us all. P(doom)=0.3.<p>Anyway, that&#x27;s why we&#x27;re working on much large LLMs. It&#x27;s important that we do this research and expand the size of the LLMs because a larger LLM could destroy us all.<p>Listen, everyone cosplays as doomers here because it&#x27;s stylish right now. But they&#x27;re all trying to make this thing they think is suicide. Is this a suicide cult?<p>No. It&#x27;s just people playing around acting important while solidifying their business.<p>Recently, I read on Twitter that many AI CTOs think that P(doom) &gt; 0.25 by 2030. Here&#x27;s my question to one CTO of a venture backed firm with present personal assets over $400k that believes this:<p>- A representative of mine will meet you in South Park<p>- The representative will be carrying a $100k cheque and a document outlining terms for you to pay a lump sum of $177k on Feb 1 2030 (10% compounding, roughly)<p>- The loan will be unsecured<p>- We will make the terms public<p>This is pretty close to break-even for you at terms you won&#x27;t receive personally anywhere else. Once doom happens money will be worthless. Take my money. I&#x27;ll also make the deal all numbers divided by 100 if you just want to have some fun.
评论 #38109445 未加载
matjus超过 1 年前
The most notable thing about AI right now is that it&#x27;s the new widget. The economy wanted it, the zeitgeist wanted it, and for <i>that</i> purpose no more development is necessary. It&#x27;s already reshaped McKinsey&#x27;s advice, Microsoft&#x27;s operating system and the public&#x27;s trust in the exceptionalism of art creation.<p>take Auto-Tune: before it blew up, one uncritically accepted that a great sounding vocal performance was simply that. The mere existence of the tool broke a covenant with the public -- artists could assume good faith on behalf of listeners once, but no more.<p>Similarly, AI&#x27;s chief effects are likely to be cultural first, and material second. They&#x27;ve already broken the &quot;spell&quot; of the creator. Seemingly overnight, a solution to modern malaise (choice fatigue, lack of education, suspicion of authority) has colonized the moment.<p>In this sense, one-percenters &quot;seizing power forever&quot; really have found the best possible time to do so -- I can&#x27;t recall a time where the general populace was this vulnerable, ill-informed, traumatized and submissive.<p>That the underlying tech barely works (maybe that will change, but I predict it won&#x27;t) doesn&#x27;t really matter.<p>I don&#x27;t generally support overly regulatory regimes, but in this case I think existing thinking around monopoly (particularly as it affects the psyche of the aspiring American) is sufficient to indicate <i>something</i> needs to happen
评论 #38109861 未加载
评论 #38109891 未加载
评论 #38111251 未加载