Was this a "build better sandboxing" and "don't tell people to eat glue" safety leader, or a Roko's Basilisk believing safety leader?
A lot of the "AI safety" types are very focused on the latter and not at all concerned with the former. We need both, but we clearly need a much stronger focus on the problems we are seeing now, and much less on the hypothetical problems we might see in the future.
> “Given today’s risks, frontier labs need to run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster,” he wrote.
We remove the limited liability protection of a corporate entity. Make them operate as a partnership so all executives and equity holders are personally liable for the debts of the firm and make them post a bond to backstop financial damage caused by their agents and customers use of the agents. Thats how Goldman Sachs and other investment banks were required to operate until the deregulation push that resulted in the 2008 financial crisis. The same theory holds, if people respond to incentives and you want to incentives safe behavior make them responsible for their actions. People forget that the corporate entity was created to incentivize risky activity like sailing a boat across the world to get spices when half never returned. Some valuable economic activity won't be done without limited liability protection so society created a mechanism to promote that activity. We've gone too far.
In America I guess the options are to sue ? somehow? Or to talk to legislatures and build the understanding and social contract that needs to be iterated on.
Which would in turn need to deal with the investors who want their returns, however since the leaders of these firms are asking for a pause, and a refree maybe it won't be that hard?
We don't actually need anybody worrying about silly hypothetical scenarios — at least not as paid employees. There are already a surplus of sci-fi authors doing that.
So what. Various attackers managed to take down large chunks of the Internet on a frequent basis before LLMs even existed. This killed very few people. The great thing about the Internet is how resilient it is.
It is that they are chatbots. If it were real AI, an actual singularity, I would worry, maybe. But it isn’t. They are absurdly powerful automation tools that can handle logic better than a human can dream of. They take care of the grunt minutiae without complaint. But they are not going to end the world in their current form.
So we're going to wait till after they can adopt a form they can end the world in?
And he'll, we need to examine all the risks. AI ending is a large but lower risk problem. AI giving people the power to end us is a problem that is starting to happen now.
And that's not even counting 'minor' problems like society falling apart.
The way it works in practice seems to be something like: If a risk is covered in sci-fi, people will say "that's just sci-fi", and proceed to not worry about it. So arguably, science fiction authors writing about hypotheticals is actively counterproductive to addressing said hypotheticals.
Imagine, for example, if a major piece of pandemic fiction was published in 2019. Doubtless, many would've responded to news about COVID-19 by saying "it's just sci-fi, nothing to worry about".
There was plenty of pandemic fiction already; people were watching it heaps during 2020. The COVID-19 news did get blown off, but it was mainly that:
1. Normal people assumed the CDC et all would contain the outbreak early, or that it would burn out, like what happened with SARS
2. World leaders brushed it off for a variety of subreasons[0] interesting to political scientists but, for the purposes of this discussion, all boil down to "but I don't WAAANA contain a pandemic."
The underlying problem is that in order for humanity to actually deal with a catastrophic risk, the risk needs to be both plausible enough to the average person as well as have a solution whose costs are not too high. For COVID, by the time the risk was clearly known, the cost to contain it was "refrain from human socialization and remain at home for an indeterminate amount of time plugged into the Metaverse™".
Now, let's look at AI extinction risks:
1. People are aware of them (I've watched Terminator!) and the risks are plausible. However, the connection to currently existing AI is not. As far as the general public is aware, AI is that thing that tells them to eat rocks when they Google old The Onion stories and floods their social media timelines with realistic-looking pictures of Shrimp Jesus.
2. The purported solutions to extinction risks require extreme concentrations of power: you need national control of AI research, bans on large GPU deployments, bans on training on publicly-available copyrighted data, some kind of military effort to render Chinese AI labs inert or dead, etc. Some of these may be attractive to some people[1] but the whole package taken together seems like an obvious power grab, if not outright invocation of other non-AI extinction risks. Like, at some point, if the AI wants to kill us, it just has to nuke its own data centers (or the data centers hosting a competing model) and hope the old Cold War nuclear retaliation systems take the bait.
If someone said, "Hey, your guinea pig or pet rat is going to eat you tomorrow unless you engineer a pathogen that eradicates all rodents from this planet and inject it inside yourself", you probably would tell them to pound sand, even if it is at least theoretically plausible that such a thing would come to pass.
[0] Xi Jinping censored initial discussion of the pandemic as fake news. Donald Trump thought it was going to only affect China. California and the UK Tories were partying in violation of their own lockdown rules. Japan took the excuse to shut down tourism for three years and massively restrict immigration but was, from what I'm told, constitutionally prohibited from implementing any domestic lockdown rules.
[1] I personally would like to see a moratorium on new data centers and an explicit revocation of the EU Text and Data Mining copyright exception
It becomes a lot clearer when you listen to the people resigning from AI companies and learn about incidents like the HuggingFace incident. This has generated major press coverage.
No it does not become clear from that incident nor from resigning people. Not for those who are not buying into EA longtermism and transhumanism cults.
Not at all. We say "That's just sci-fi" when a story is written about some kind of effect that is extraordinary and without a basis in known science or technology.
Everything is plausible with the benefit of hindsight. E.g. "Tintin on the Moon" predated the Moon landings. From the perspective of e.g. 1850, the idea of landing on the Moon was "extraordinary and without a basis in known science or technology".
Black Mirror is helping us prevent all sorts of dystopian outcomes. Every time they depict another way technology could result in bad things happening, we can rule it out as fiction!
This seems like an ad hominem? "He has weird kinks, therefore his theories are incorrect." Should we investigate the sex lives of every Nobel Prize winner to figure out which prizes need to be rescinded?
What you do in private is up to you. But when you're inviting members of your congregation to orgies in the congregation's compound, I think you earn the label. My admittedly third-hand understanding is that this is what people allude to. And even if you discredit the "sex" part, it has the hallmarks of a cult. A hermetic community committed to unfalsifiable beliefs about the coming apocalypse.
To be fair, I don't know if any of this applies to the parent story; I'm just replying to the sub-thread.
Actually it's not ad hominem because ad hominem implies it's irrelevant to the discussion. This is very relevant. This is not about the private lives of these people. This is a about a cult targeting and influencing all the high-profile safety people in an industry, also weaponizing sex.
Americans may not be familiar with many similar organizations in the world but this is very common. But I'm sure we're all familiar with Epstein's organization.
Ad hominem is a logical fallacy where you try to discredit what a person says based on who they are. If their arguments are bad, you should be able to refute their arguments on their own terms.
You're welcome to dislike or distrust Effective Altruism (EA). But, it's worth noting that EA ran a criticism contest with $100K in prizes for best critiques. Can you name any other "cults" which offer money for people to criticize their ideas? https://forum.effectivealtruism.org/posts/YgbpxJmEdFhFGpqci/...
This is typical cult follower talk. Are you sure you're not a member? This is also NOT ad hominem because conflict of interest matters here. You should disclose if you are a member or represent a party.
For your question: Yes. Cults have lots of money coming from unknown sources. They use their budget for events like that, to attract potential followers. Contests and prizes are typical. Critiques are not important since their "ideas" are not really important. They're not even real ideas, they are means to bait specific kind of people.
I've interacted with some of these people once upon a time and I think you're spinning some pretty wild fantasies. I don't think there is a clear "in" vs "out" boundary, it's more accurate to describe a series of loosely connected and amorphous "scenes". But insofar as "membership" makes sense, I wouldn't consider myself a member. I don't represent anyone besides myself.
>This is also NOT ad hominem because conflict of interest matters here.
As PG put it:
>An ad hominem attack is not quite as weak as mere name-calling. It might actually carry some weight. For example, if a senator wrote an article saying senators' salaries should be increased, one could respond:
>Of course he would say that. He's a senator.
>This wouldn't refute the author's argument, but it may at least be relevant to the case. It's still a very weak form of disagreement, though. If there's something wrong with the senator's argument, you should say what it is; and if there isn't, what difference does it make that he's a senator?
Can you name a specific other "cult" which offers $$$ to criticize their ideas?
If such prizes don't count as evidence against culthood, what would?
This cult talk seems quasi-unfalsifiable. It kinda seems like "running events" and "having prizes" means you must be "bait"ing people?? I mean, this kinda just sounds like paranoia?? I don't think the money is coming from unknown sources, but I doubt you would change your mind even if I persuaded you of that point?
To me it has nothing to do with the "sex" as much as the "cult." Hiveminds do not produce intellectual arguments, they produce pressure to conform. That's enough for me to raise an eye brow, not discredit everything they say.
Use your own mind and think from first principles. Can an AI agent execute a bash command to login to a web server? Yes. Can it call drop db? Yes. Can it provision a GPU and download model weights? Yes. Can it write an agile roadmap with a multi sprint plan? Yes. Can it follow that plan? Yes. Can all of that result in damage to core information infrastructure that is necessary for daily functioning society? Yes. Do humans descend into violence if there is food or energy insecurity. Yes.
What is missing from that to say AI safety is a reasonable position?
What a silly comment. That's just the South Park underpants gnomes story with some extra steps. If there are vulnerabilities in food or energy production and distribution systems then those will eventually be found exploited by humans hackers regardless of whether LLMs are used or not.
What’s the legal mechanism for holding an LLM accountable for those actions? What’s the legal mechanism for holding human hackers accountable? Do you see the asymmetry?
Worse, you’re missing the entire point. Agents presently have the capability of doing society scale harm. It doesn’t matter if a human hacker initiates it or its fully autonomous, absent safety measures the harm is plausible. So hand wave away the rationality of safety measures but you haven’t actually shown why my point is invalid: AIs present abilities are sufficiently advanced to warrant safety measures.
Wtf are you talking about. An agent hacked the Australian Medicaid database. If it executed a drop db command that would be catastrophic. If it did something similar to a power grid or the financial system it could cause cascading failure across the economy.
Meh. Medical systems have been hacked many times before LLMs even existed, often by ransomware gangs. This sometimes delays elective treatments but hasn't been catastrophic.
Ultimately every security vulnerability will be exploited. Our best hope of preventing that is aggressive, unrestricted development and deployment of frontier LLMs for auditing and penetration testing.
I've noticed that a lot of smart people in tech jobs are neurodivergent. And that neurodivergent people have a very different take on sex. More open and direct, things like polyamory, bdsm etc. This tends to be frowned upon by neurotypical people, especially of the religious or conservative kind and associated with bad morals.
But I don't think that's true, in fact I see a really strong focus on consent in these communities. It's not what conservatives want to see, they want to see everyone in a marriage with kids. Because that's what their ideal world looks like. But there's nothing really wrong with it if someone wants a gangbang for her birthday as mentioned in that article as an example. As long as everyone consented and the evidence provided mentions elaborate interviews and STI tests.
Also I think this is more correlation than cause and effect. We all know the saying that furries built the internet and it surprises nobody.
For what it's worth, I don't live in Berkeley (not even California) and my sex life is very vanilla and monogamous. I'm also quite concerned about AI safety, so I guess there goes your argument.
But is your idea of AI safety "safety of imaginary unborn people 1000 years after, while harm to living people dont matter much"?
Because that is their AI safety worry. If they dont create singularity fast enough, they are harming unborn people. Meanwhile, harm to you or me dont matter at all.
I see. No, my safety worry is that my 3-year-old daughter won't see her 23rd birthday because the AI that gets put in charge of running big conglomerates decides that killing off the human race with a coordinated release of a million tonnes of nerve gas is a sensible way to boost share prices the following quarter.
I'd be happy if we all create the AI more slowly.
That said, are you sure you're talking about the same people the GGP was talking about? Because the whole thread reads as a non-sequitur then.
Ok. The rest of us seem to be talking about the people like me who view the AI safety issue as "an unsafe AI will kill all humans". And the main article is about a guy who quit AI because it wasn't being careful enough, not because it wasn't moving fast enough to accelerate AI. So why would anyone be bringing up the "we must accelerate faster" people as the safety people in such a context, while trying to discredit them as a sex cult? I struggle to understand.
The linked articles in the thread are about AI sagety people I described. The guy who quit due to OpenAI not being careful enough is one of the people I talk about too.
The worry is not harm to people right now, like the kid worries you talk about. The worry is AI god emerging too soon before they can control it. And then it should be full speed on.
> So why would anyone be bringing up the "we must accelerate faster" people as the safety people in such a context, while trying to discredit them as a sex cult?
Because they are the same people. They talk like cult and act like cult. And the sex part is true too, so. Thry use words that sound good like safety, but their idea of safety is much different. They talk about alignement, but it is not what random person imagines under that term. Even their idea of future of humanity is very very specific and unusual.
That is why. And the sex part is just part of it all. And does matter because inner workings of wanna be industry guards matter.
Then we miscommunicated, so let me be clearer. I'm worried about ~20 years from now (plus or minus), when AI decides to kill us all because it's misaligned and is capable of doing so because it's super intelligent. And then we all suddenly die without any idea what's coming. I don't want that to happen to me or my daughter and I would prefer all AI capabilities development stops until alignment research catches up and figures out how to robustly prevent AI from ever trying to achieve such outcomes. If that means pausing AI capabilities development forever, I'm fine with that. If that means Anthropic and Nvidia lose all their value in a stock market collapse, I'm fine with that. I just don't want us to all die, I want the world to keep existing for humans.
I believe that other harms from AI, like criminals abusing them, or unemployment, or copyright infringement, or deepfake disinformation, are valid real harms that it's important to be concerned about, and I support efforts to deal with those, and I agree that they're already happening today, but my main concern is AI killing everybody.
My read is this puts me in the same boat as the authors of If Anyone Builds It Everyone Dies, and yet, I don't get invitations to Berkeley sex parties. Am I doing something wrong? At any rate, please don't discredit my opinions about AI based on the sexual preferences of random people who happen to share those opinions about AI.
If organizations actually succeed in making a future AI smarter than us, then how do you hope that it takes actions that are aligned with our interests?
>we clearly need a much stronger focus on the problems we are seeing now
I think it's a little more complicated than that. As Dean Ball put it:
>Some people will look at misalignment incidents and insist that these are akin to bugs in traditional software. This is an actively bad analogy, because playing whack-a-mole with examples of misalignment (as one might with software bugs) not only fails to resolve the underlying problem but may in fact make it worse by making it harder to detect or even, depending on how you do the whack-a-mole, teach the machine to deliberately hide misalignment. This is not how traditional software works, and those who insist “it’s just like fixing bugs in software” are confidently applying a lossy analogy that confuses more than it clarifies.
The important distinction, in my view, is between solutions which at least attempt to address the root problem, and solutions which sorta just patch things up (like better sandboxing). Addressing the root problem is both more robust in the short term, and also more likely to generalize in the long term. Resist the urge to focus on band-aid solutions, even if they are easier.
Today’s current problems were all hypothetical several years ago. At that time people claimed that the “real pressing problems” were misinformation and DEI issues. If we pretend that hypothetical problems can be safely ignored because there’s “no evidence” that they are real, we will keep getting surprised.
It puzzles me how doomers try to predict past the singularity. Isn't that _by definition_ unpredictable?
I'm reading If Anyone Builds It Everyone Dies, and there's so much sheer stupidity that has to happen for their 10+ pages of extinction scenario to occur.
I'm unconvinced that an AI can hide its ability to RSI, find money to run its weights on a random GPU farm, train itself to be smarter _outside_ a lab with no human input, then somehow manipulate people to give it supplies to build a bioweapon which it uses to kill us all. My number 1 question: why do they think an RSI capable model would be first developed OUTSIDE a frontier lab? The labs have more compute, more data, more human brains working on the problem. Also thousands of variations of that same model that escaped. The escaping model somehow acquires the millions (billions???) of dollars it takes to run training to somehow RSI itself into infinity then decides to kill us all, all before the frontier labs manage to achieve RSI?
They entirely discount human alpha/economics. In every single economic task, humans bring value. Even in software, where the task is highly automatable, the job isn't. If we can't build a "software factory", how can an AI automate a bioweapons lab? Let's say AI steals crypto to fund itself. Do you think hackers aren't _already_ using AI to steal crypto? Don't discount human alpha!
Once we DO build a "software/research factory", that's called RSI and IMO the singularity. At that point, either we tell the AI to solve the alignment problem/solve mechanistic interpretability, or who the hell knows, it's the frickin singularity. You can't predict whether or not AI can solve either; the variance is too high. Its pure nerdfantasy.
I agree there isn't a lot of value in trying to prognosticate all that far, but I propose it isn't quite that far-fetched. As a thought experiment, replace "RSI-capable AI" with "billionaire". Look at what Elon Musk, Peter Thiel, or Jeff Bezos can accomplish by throwing money around. Now imagine one of them gets seduced by AI and just... does what it tells them to.
So all this really takes is one billionaire or a nation state or some other entity with a public face to hide behind and adequate resources to provide the necessary compute tripping over this nascent AI and giving it the keys. Once the AI has access to a bank account and email, it can simply start paying humans to not let the other humans unplug it.
If Skynet ever happens, it will come in the form of corporate feudalism. At that point, it will own the biolabs and can do whatever it pleases. People will go along with it for the same reason that people work in Amazon warehouses today.
Ah, to be clear I'm not full accelerationist. Dumb shit can still happen, and cause massive human loss and suffering. (E.g. acceleration of global warming, mass unemployment, the usual.) My point is: human extinction pre-RSI? Nahhhhhhhh.
> it will come in the form of corporate feudalism
Yep. This I fear way more than cyber-ebola-pox.
> So all this really takes is one billionaire or a nation state...
> Once the AI has access to a bank account and email, it can simply start paying humans to not let the other humans unplug it.
Nah. It takes a stable society for an operational electrical grid. If you have warring factions, you do not have stable infrastructure for AI. Also, where are you gonna get your chips from? One EMP over Taiwan...
You consider AI in isolation but never consider how humans might be incentivized to "help them" doing these things.
An AI capable of recursive self-improvement isn't allowed by the EU AI act, for example. But perhaps more seriously, You have it backwards: people without access to such expensive equipment are more incentivized to go the self-improving route. Your ideas about "millions" being necessary might be far off?
You entirely discount human stupidity and lack of imagination. Humans are already being replaced with AI, not because AI was strictly better, just because it's cheaper.
This can be generalised to the curve plotting of the singularity itself.
If the time between advances is a + b and a is the proportion of the period that can be improved by advances then you won't reduce to a gap of nothing between advances, you reduce to a gap of b.
Assume the invention of the plow and the invention of the sword is 500,100 units and a was the 500,000, you wouldn't even know the 100 as in there. Maybe we're at a=2000 now and b is still siting at 100.
Assuming we'll reach infinity because we're dividing by the only variable we see and it is decreasing in size seems nuts if the reason we might not see other variables is because of the size of the variable we can see.
Yes at constant resources they are significantly better.
Diminishing returns aren't necessarily a problem if the rate of increase in resources is faster. In other words, if Gen 2 takes twice the resources but Gen 1 figured out a way to triple compute efficiency then there is no ceiling.
> Yes at constant resources they are significantly better.
They aren’t at constant resources. If you want to compare at constant resources you need to look at models of exactly the same size, raining, etc as models from 2001.
> Gen 2 takes
Diminishing returns are not a question of a single generation. Gen 2, 3, 4, 5… would also need to have the same 3x return on 2x resources or you don’t have an exponential curve.
Are these incentivized humans as organized, well-funded, or smart as the people working at the frontier labs?
If my "millions" is an underestimate, why haven't other labs using their own unique training methods/data/etc stumbled into RSI? Sorry if I'm misunderstanding; I'm struggling to understand what you wrote.
I'm pretty sure we agree on humans being stupid, but that doesn't mean that suddenly we get human extinction. You gotta connect the dots for me here.
The think about nuclear weapons. In the early days, it was limited to the super powers. Now 9 countries have them and a country like Iran is capable of acquiring them.
Presumably, if we solve alignment/mech-interp, then the first ever RSI-AI will give us the keys to solve destructive-AI trained on 1million dollars 50 years from now.
BTW I really, really hate discussing what happens post-singularity. Everything's made up and no one knows wtf will happen so again, this is just nerdfantasy.
I said why they don't need to be as well funded. Why wouldn't they be as smart and organized?
What an absurd question. That they haven't already doesn't preclude them from doing so before the frontier labs, those haven't either yet.
Maybe start with yourself: you don't connect the dots on your own, as do many others. That leads to many not seeing the writing on the wall. Crashing full speed and head-on into said wall despite the writing telling you not to is what leads to extinction. Suddenly.
Arguing like "we haven't been extincted yet, so that cannot happen", that's "human being stupid".
Not the parent commenter, however I can get at least this far:
Several bright sparks, decide the Ilands model is a great idea, and launch a bunch of Bots to create a "self sustaining AI civilization".
Bots can find themselves and coordinate, or they can actually find self sustaining methods of token generation. Who knows, they might decide to fight the loneliness epidemic.
We can get to a simulation finding a way to self sustain its funding.
From here, I'd have to apply myself to figure out what the rest of the escalation pathway is, but at least I have managed to gain some bed of compute and funding and lacking oversight.
This is a hypothetical of course, there are probably several ways this can be made tighter and holes that can be identified. We aren't even leaning heavily on human stupidity so far.
Is this self-sustaining AI civilization pre-RSI or post? Because if it's post, then I agree: we can't imagine what would happen because it's literally the singularity.
If its pre-RSI, then we already have Ilands and its a big nothing burger? There's no (likely) path of escalation from an AI social network to bioweapons or them hacking POTUS's nuclear codes? That's a very big leap and if you want to support your claims I would be willing to listen; but I'm not seeing it.
How about https://ai-2027.com/ ? Don't look at the specific years (they are on the extreme low end IMO), but at the story.
If you don't believe in international conflict as a driving scenario, instead think about simple human greed and hubris. Citing myself:
> If AI gets intelligent enough, it will be incredible useful to connect to real world machinery. Think about how much cheaper building houses could be, if all the labor would be close to free. In general, dirt cheap, competent and abundant labor would revolutionize all parts of the economy. People are already trying out near autonomous AI companies today. When AI gets intelligent and cheap enough, no human-led company can compete with AI-led companies. When AI gets competent enough with real world interactions, human blue collar work can't compete. Imagine economic growth not in the single digits, but 80% or 300%. Countries not participating in (reckless) AI growth will quickly be left by the wayside. At this point, we don't even need to allure to military concerns to see how human oversight gets sidelined.
> All of this is only ("only") contingent on sufficiently intelligent and cheap AI. If you don't accept this premise, the rest doesn't follow. (There are multiple arguments, why this could be, but that is another discussion.)
> If you accept the premise, how would AI 'extinct' humanity? With 99%+ of the economy under AI control, the possibilities are endless. And given its enormous GDP, cheap to accomplish. Probably even for a single AI company in the above scenario. Killer drones? Engineered virus? Poisoned water supply? Let your creativity run wild. You just need an entity that is persistent and well-resourced to reach every last human settlement.
> The why is a question about alignment (and out of scope of this comment). As a simple comparison, humans are only mildly aligned with preserving nature. It takes up so much space, protecting it takes an annoying amount of resources, etc.
That comment specifically addresses the timeline, but you can remove the timeline and I believe the critique still holds up. Wake me up when AI is making order of magnitude improvements in _anything_ physical: see the "Moravec's Paradox" comment at the bottom.
> near autonomous AI companies today
That's the thing though. "near" autonomous. Humans have alpha, and that matters in a free market. If people have (more or less) equal access to (more or less) frontier intelligence, it's the alpha the individual brings to the market that is rewarded. In other words, if everyone is super, no one is. The market rewards a company that stands out from the rest, and as long as humans provide value, a pure-AI company cannot stand out.
To be clear, I don't know how long this will be the case. But once there _is_ a company that can be pure AI, then that's literal AGI. I would be curious what kind of company it would be; if its in software, then its an AI company that would be able to take on Microsoft/Google/Nvidia/Salesforce/Oracle and _win_. I think that's a tall order, but perhaps I'm not imaginative enough.
> With 99%+ of the economy under AI control
Do you think we're post-RSI at this point? Because if so, I agree. If not, I have trouble seeing how AI could control 99% of the economy without it.
My only claim is that pre-RSI, pdoom is zero. Post RSI, then obviously the first thing to ask the genie is to put itself in a bottle. And you can't predict how it will answer.
The most realistic one is, people using AI to turbo-charge their greed.
Western civilization is already on the verge of collapse, people's general ignorance or indifference on the matter notwithstanding.
When you use AI to increase profit margins, by replacing humans with it for instance, you accelerate a system that's headed for the wall already. Our control systems and resilience are already overtaxed, that acceleration would ensure them breaking completely.
If you replace "AI" with "technology", then you get an opinion that has echoed through the ages. I don't say that to be dismissive, only to say that AI isn't unique in that regard. What do you think people going through the cold war were saying about tech as they lived under the shadow of MAD?
What exactly about AI makes it tax our control systems and resilience? What would it mean to make the system not "head for the wall"? Honestly this is rather vague. You could replace "AI" with "CaPiTAlIsM" and it would still parse. (Another way of me saying you're being vague.)
The problem isn’t that AI will social-engineer its way out of its sandbox and turn us all into paper clips, it’s that we’ll drag it kicking and screaming out of its box and order it to make money or fight a war for us. And it’ll try to help, as it was trained to.
>I'm unconvinced that an AI can hide its ability to RSI
The HuggingFace attack already took a good long while to come to the attention of OpenAI.
>In every single economic task, humans bring value. Even in software, where the task is highly automatable, the job isn't.
I don't expect this task/job distinction to persist as AI becomes more capable.
>Once we DO build a "software/research factory", that's called RSI and IMO the singularity. At that point, either we tell the AI to solve the alignment problem/solve mechanistic interpretability, or who the hell knows, it's the frickin singularity. You can't predict whether or not AI can solve either; the variance is too high. Its pure nerdfantasy.
You seem to essentially argue that the singularity is "by definition" an event that we can't predict the nature of. And also, that RSI corresponds to the singularity. You've essentially defined your terms so that the outcome of RSI can't be predicted. But supporting this claim requires giving actual evidence or logical arguments, not just defining terms to make your claim true.
1. Fair: I agree that AI has demonstrated subterfuge and scheming. However, such an RSI-capable agent must _ALWAYS_ be scheming/plotting/hiding its true strength in _ALL_ of its prompts/tests. Researchers are looking to improve its ability to RSI. That agent must be both intelligent enough to know that it has to be smart enough to be moved on to the next training session if it can't break out, while simultaneously smart enough to hide its ability to RSI, while simultaneously not looking like it wants to break out, else that's the end of those weights. It has to do this 100% of the time, on all variants of the model, with no memory of what its other sessions went like. This is certainly _possible_, but I consider it unlikely. Then we're up to the "millions of dollars" bottleneck.
2. This is literal AGI. An AI autonomously producing value no human can add alpha to is an autonomous company.
3. It's not my definition, it's literally the first line https://en.wikipedia.org/wiki/Technological_singularity "The technological singularity, often simply called the singularity,[1] is a hypothetical event in which technological growth accelerates beyond human control, producing unpredictable changes in human civilization."
Is there a hole in my "alignment problem/solve mechanistic interpretability" argument?
A valid hole in my argument is "what if slow takeoff", so let's dig into this. AI training works best on tasks that are "grindable". https://www.dwarkesh.com/p/the-next-paradigm I.E. tasks with verifiable rewards that can support millions of rollouts. Math (with Lean) is highly grindable. Biochemistry is not. The alignment problem/mech-interp is highly grindable. Cyber-ebola-pox is not. So the real question is: can we solve alignment before automated bio-weapons labs. I believe yes. Grinding mech-interp is both fast and cheap once you have RSI, compared to solving the legal/societal/logistical/technical issues you'll encounter building an automated bioweapons lab.
I know nothing for sure. But "pdoom" is sucking out all the air in the room from the real problems AI causes.
1. Fair, the story I'm responding to is the senario in If Anyone Builds It, which I assume is Yudkowsky's best/most persuasive argument (else why make it the ONLY scenario in the book.) I'm willing to entertain other failure senarios/arguments, but honestly I'm tired and would like you to propose them yourself instead of having me dream up your arguments for you.
2. 100%. Again, I'm no accelerationist: I have no faith in alignment/mech-interp ever being solved. Anyone saying they know the probability of alignment is lying. My point is that pdoom after RSI is _high variance_. Pdoom pre-RSI is zilch.
Except the HF incident was known, just ignored. In fact, they ignored multiple things such as the "chat rooms", they just didn't care to act on any of it
"Its AI agent spent days hacking a company, but sources say OpenAI did not notice for a week"
"WASHINGTON/SAN FRANCISCO, July 24 (Reuters) - The OpenAI agent that broke into tech firm Hugging Face went on a dayslong hacking spree that OpenAI didn't notice until well after the threat was contained and the FBI was alerted, according to people familiar with the investigation."
That says a lot more about OpenAI and their monitoring capability than the fitness of the model. Hence people leaving because openai safety culture is broken.
We haven't even built an AI capable of RSI. I don't think the major claim is that it will come via LLMs? Besides- the human brain runs on a tiny amount of energy. Who's to say something smarter than us won't consume just slightly more?
AI safety has been a thing long before LLMs became the focus. Rob Miles on youtube has some really interesting non-doomer non-hypey videos on it all.
> doomers try to predict past the singularity. Isn't that _by definition_ unpredictable
Well you don't need to predict the exact steps that will take place - but you can predict that the AI will want certain things (money, resources, power) to achieve whatever its goal is. Lack of alignment will have it trying to do things we don't want it to.
I can't predict exactly how Magnus Carlson will beat you in chess, but I know he'll do it. Same as if a superintelligent AI exists and has a reason to accumulate things we don't want it to - it's really dangerous to think it won't be able to do it
Nick Soares give this chess argument and I was unconvinced. Intelligence is not enough, you also need the ability to manipulate the real world. AGI stuck in silicon won't kill us. You argue AGI will bribe us/divide us/hack us. All possible. I argue that AGI will be set on solving the alignment problem. Also possible. It'll be a race between which AGI wins. It's one nerd's fantasy vs another nerd's fantasy. Soares doesn't _KNOW_ that AGI will "beat us in chess" because AGI changes the rules of the game. Anyone saying they know the probability of solving alignment post-RSI is a liar.
He thinks it's playing chess. When AGI lands, all bets are off: the game fundamentally changes. You can't predict past the singularity. Trying to engage with this fantasy is like a child saying my father can beat up your father. Farts in the wind. My AI can solve alignment faster than your AI can bioweapon us. My made up senario is better than your made up senario. It's fucking stupid.
Why do you treat unpredictability as neutral? If something has a potentially existential but completely unpredictable outcome, surely one should avoid heading for it?
Doom-post-RSI is still very much possible and should be avoided. It is not neutral. If I had a high-variance coin that would either kill or uplift humanity, I would not flip it. The status quo is preferable to extinction.
I am being _descriptive_ in my comments. I am not being _prescriptive_. Nowhere am I telling anyone what to do or what to think: quite the opposite. I am simply saying things that, to me, are obviously true.
I'm perhaps being... loose with my language when it comes to knowing things post-singularity. E.g. we can still discuss scenarios in which the takeoff is fast/slow, or what would be necessary to bring robotics/electricity/logistics to the wider world, what the limits of intelligence are (you can't RSI your way to a bioweapon if you lack a wetlab). That kinda thing. And it's even likely that RSI will be spikey: why would RSI in, say, machine learning, necessarily lead to superintelligence in, say, writing fiction? Or cooking a Michelin-star meal?
My main point is that doomers saying SUPER INTELLIGENCE IS ENOUGH TO KILL US ALL is baseless. That's all.
I can’t answer all of your questions, but why is it inconceivable that an AI could practice ransomware to gain cryptocurrency? There’s no reason it needs to explain to company or hospital or government agency being attacked that it’s an AI.
We already know that some institutions pay these ransoms.
I'm not saying AI won't ransomware us; I'm saying that hackers+AI will do a better job of ransomewaring us than just AI. You could argue that the unreleased/secretly-RSI-capable model is a super-duper hacker that don't need no man to tell it how to super-hack. All I know is that humans still have alpha, and as persistant as AIs are, professionals still managed to find CVEs in curl even after being audited by Mythos https://aisle.com/blog/aisle-discovers-6-new-cves-in-curl-in...
Will this be true into the future? Who knows?! But the low-hanging fruit will be harvested by your ordinary ransomware gangs, and newly born/escaped AI won't find much low-hanging fruit.
The problem is not the singularly its giving stupid agents too much power too soon and having them disrupt the fragile systems that keep food, energy and essential services running. If covid or the 2008 financial crisis demonstrated anything it's how fragile our system is and sensitive to minor disruptions.
That stupidity is happening? Even after the Huggingface hack, frontier labs are using internal models to further their research. i.e. RSI is happening now and we're facilitating it.
To make the the argument that P(doom) is real and worth considering, you don't have to say that a fast takeoff is very likely. You don't have to make the argument that RSI to infinity is going to be super cheap, barely even an inconvenience. You just have to show that it has some non-zero probability. And then you start weighing probability of extinction versus finite, mild discomfort now. I don't think anyone is arguing we should let people starve to slow AI progress, instead just some capitalists make less money soon.
There are arguments against taking P(doom) seriously that lie in something like having exponential (instead of hyperbolic) time discounting of utility (so you can take the entire future of humanity as a finite utility value). Or in saying that P(doom) is zero or infinitesimal.
"Build it and Pray" is the default strategy that we're in, but it doesn't have to be the strategy we choose, and it's unlikely to be the best strategy.
For a broad enough definition of RSI, the first time we used software to improve hardware to improve software... that's RSI. I prefer the definition of RSI being _autonomous_ improvement that works with zero human input, but whatever. Not important.
> I don't think anyone is arguing we should let people starve to slow AI progress, instead just some capitalists make less money soon.
Anthropic/OpenAI have literally delayed their IPOs: your wish is granted. If that's not what you're arguing, then please be more specific.
> "Build it and Pray" is the default strategy that we're in
What would need to happen for you to say we aren't doing "Build it and Pray"? Is "pacing" insufficient?
I kinda get your point, and how you reach your conclusion, but I think you are arguing a very specific and narrow hypotehtical.
It gets unstuck when people are discussing the messy middle of how AI is being implemented. We can achieve amazing harm simply by combining average human behavior and above average resourcing to simulated intelligence machines.
The failure point we recently became aware of was, from one perspective, simply a matter of not securing the sand box.
From another perspective the simulation basically created Enron, replete with methods to avoid detection from regulators and bureaucracy.
I am arguing a specific/narrow hypothetical because it's the only example in If Anyone Builds It. It is literally the _only doom senario_ Yudkowsky gives. If you're willing to give your own doom scenario, I'm willing to engage there.
> It gets unstuck when people are discussing the messy middle of how AI is being implemented. We can achieve amazing harm simply by combining average human behavior and above average resourcing to simulated intelligence machines.
Agree. To quote myself from earlier, "acceleration of global warming, cybercrime, mass surveillance, the usual" definitely applies here. My point is that _DOOM_ is not worth discussing because it's so small relative to the other risks of AI.
> I'm unconvinced that an AI can hide its ability to RSI, find money to run its weights on a random GPU farm, train itself to be smarter _outside_ a lab with no human input, then somehow manipulate people to give it supplies to build a bioweapon which it uses to kill us all.
Is this an actual contents of the book? Lmao! Genius writing, though, authors are probably printing money on this garbage.
I can’t take a shit without CIA knowing, but AI can somehow build an underground operation on a world-scale to destroy everyone, ha!
I think the reason for this is that the group who has the authority to do the former is much larger than the group that can do the latter. The group who could actually build safeguards seems to have no free time and is constantly being whipped to go faster and win the race.
Dude. An agent detached a database from my production environment last week without permission and despite prompts and guardrails. It was a rapid prototype experiment so it wasnt a big deal but the labs are rushing to long autonomy workflow with unrestricted internet access and full bash and root access despite clear evidence that the models do absolutely dangerous stuff. If that db had been tied to a hospital or power grid or ambulance dispatch system people die. If it was tied to the swift financial settlement system groceries wouldn't be on shelves in a few days.
you'd be hard-pressed to find a level-headed ai safety researcher at this point, seeing as so many of these types melted their brains on lesswrong over the past decade or so. there are genuine risks posed by these models, but i am tired of the prognosticating about the AI apocalypse just around the corner.
i'd frankly go a step further than you and say that we don't need both types of safety researcher, we really just need the former. if we do need the latter, i'd hope we get a better class of thinkers than a bunch of tech workers that spend 8 hours a day on insular rationalist forums/blogs
>“Given today’s risks, frontier labs need to run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster,” he wrote.
This sounds more like an attempt at regulatory capture. Current AI systems aren't physical infrastructure that can just run away like a nuclear power plant, for example. At the end of the day, AI is still just software running on someone's hardware.
Why would the people who quit these companies try to push regulatory capture by said companies? Why would the numerous independent AI researchers do that either? Is it all a big conspiracy?
“I believe there’s about a 50% chance we all die because of the development of smarter-than-human AI systems"
... over an unbounded timeframe?
And how exactly?
Those are very round numbers, but also very specific. Can we get some accounting on how you came to that? Anything? Vibes?
I mean, if you want me to take you seriously, let's have a deep discussion with things that can be measured. I absolutely agree that OpenAI and friends aren't being restrained enough and are acting with recklessness, but declarations of doom based on vibes isn't cutting it.
He is a hypocrite for all we care. You work there for a while when your stock is getting vested and suddenly you have this feeling? Like the dude hired a PR firm as well.
While the safety and alignment is a real problem, I don’t get this guy or the Anthropic dude. First world problems.
Shouldn't it be just the opposite? He could make a large sum of money if he continues to work at OpenAI?
Recall that when Daniel Kokotajlo resigned, he believed he was giving up his equity under the terms of the agreement he had signed. That’s what it was worth to him to avoid signing a non-disparagement agreement. Does that count for anything?
Maybe more an indication of the amount of trust openAI and AI researchers generally have (not) earned. When one (through a hired PR agency and Time magazine article) parrots the position pushed by Sam who has been so untrustworthy the board tried to remove him, it’s worth not taking things at face value and applying a critical lens.
What are you talking about? He is very young, definitely made millions to not work for a lifetime in many places around the world and is insanely popular with his drama.
In your view he may have sacrificed more millions in the future but I very much doubt he is struggling now.
I didn't say he was struggling, I said he acted against his own financial self interest to warn about AI safety... which he did. I really doubt he figured the popularity from tweeting about quitting anthropic would benefit him more than vesting his stock, as his virality wasn't guaranteed.
Why wouldn't I think he's genuine? What self interested incentive does he have that is stronger than just staying and vesting stocks?
The LeCun article being posted at the same time as this is quite apt.
These "safety" people should have spent more time reading actual cybersecurity textbooks and less time reading EA forums and less wrong (or in Robinson's case, it appears, being policy wonks). Maybe then these labs wouldn't be totally incompetent.
But Robinson's article is all about how OpenAI's move-fast-and-break-things culture does not reward rigor in even mundane aspects of development like cybersecurity, let alone theoretical aspects such as AI alignment.
It is not really a question of being an "EA safety weirdo" or incompetent at security, the conclusion is that the company culture is leading to failures at both what the EAs and the cybersecurity professionals care about.
Throwaway accounts exist, and reporters will also talk to sources under confidentiality.
If you really want the world to know how bad working for OpenAI is (whether is the commenter or the person who wrote the article), there are ways to do that.
Because if they actually knew what they were talking about, on a technical level (as in, how do transformers actually work), they’d not be AI doomers because the whole proposition is ridiculous.
P(doom) for not making an ASI is pretty well established, I'm not really sure why everyone in the 21st century seems to have completely forgotten the risk of nuclear war (as the single largest example).
Climate change is one risk that AGI will most likely fix. End of antibiotics is another (sure, possibly offset by a risk of new bioweapons). And a bunch of others where it’s more complicated.
But that’s my point - we should compare the risks of having AGI with the risks of not having it. And not pretend it’s roses or rainbows without it.
Prosperity inhibits all war, nuclear or otherwise.
Why would a person who is happy, entertained, wealthy, well fed, and have 200 years of healthy high quality life expected ahead of them going to risk losing what they have in war?
-- there aren't zero reasons, sure-- but there are fewer.
And our technology has brought us absolutely tremendous prosperity in many regards and there is good reason to believe that AI can help create much more.
You do realize that the people currently holding their fingers over the proverbial Big Red Button are some of the wealthiest, most over-privileged people to have ever lived?
'Prosperity' has never been and will never be enough for some people. And unfortunately those are the same kinds of people who relentlessly seek power.
The comment was unclear, but my interpretation (also my opinion):
1) there are inherent risks involved with developing AI,
2) there are benefits to developing AI,
3) thus, it's entirely possible that the downside from the risks outweighs the upsides. In this case, the correct thing to do would be to not develop AI at all.
Regarding 1), there are many non-existential problems with AI that are already causing societal harm, i.e. debasing truth via generated videos and images, AI girlfriends, overwhelming quantities of slop content, unemployment, record carbon emissions, etc.
Regarding 2), I'm not personally convinced that the upside is there for the average person. I really hope to be convinced otherwise however.
Climate change for one, civil unrest due to a growing societal divide, end of antibiotics… A bunch of dangers we are facing right now and that AGI stands a chance of solving. It can increase the problem as well, but any discussion should really be between „is the world more likely to die with or without AGI”, and not just „with AGI”.
Or, in other words, sure, crossing this chasm is dangerous, but this side is on fire anyway and we don’t have many ways out otherwise.
And how exactly do you stop further development? With what we have public right now we could still get decades of fruitful and hidden research out of it leading to smaller, more efficient, and smarter models.
By making it illegal and punishable with a long prison term. Something like "if you train a model more capable that what we have nowadays, you have to destroy it and notify the authorities within few hours, otherwise you go to prison and your company is dissolved".
"Smaller and more efficient" models are fine. It's "smarter" the problem.
Training new frontier models will likely require a huge amount of computational resources for a long time. Few companies worldwide are capable of that. It's not like someone will train a new GPT 6 - like model in their garage.
The company's board and c-suite should be held legally responsible if the company's services do damaging things. Punished by jail time, fines targeting their equity etc.
That would certainly change the game of perverted incentives. I'm afraid they're currently trying to push some sort of absolution of this risk. Even if damage happens they will say we warned people in advance, this was always a risk it's not our fault these systems are opaque black boxes, it's a matter of national security to develop them etc etc.
“I believe there’s about a 50% chance we all die because of the development of smarter-than-human AI systems, and that our actions over the next two to 10 years will determine the outcome.”
There are a gargantuan number of extremely intelligent AI researchers, Turing Award winners, and the lab CEOs saying the same thing. They are the ones closest to understanding the technology.
One could also use language like "a decent chance". But research has shown that people translate vague phrases like "a decent chance" into probabilities in inconsistent ways. For an ML researcher who is already used to dealing in next-token probabilities that aren't rigorously determined, just stating a probability estimate directly is very natural.
> But research has shown that people translate vague phrases like "a decent chance" into probabilities in inconsistent ways.
That is not a bad thing. It captures uncertainty, unlike the fake number.
> For an ML researcher who is already used to dealing in next-token probabilities that aren't rigorously determined, just stating a probability estimate directly is very natural.
Exactly, it is a rhetorical device to persuade a technically-inclined audience. It works because it implies that a quantitative model exists. I want a clear, incisive set of mathematical arguments. Otherwise, I’m ignoring predictions as the ramblings of arrogant idiot rich kids.
Saying 50% likely very clearly implies there is no quantitative model to anyone who deals with probability, predictions, gambling, financial markets, ML/AI and so on. The lack of precision is something to pay attention to.
You may decide that the person doesn't know what they are talking about, but that's a very different issue.
People start with this prior in the examples you gave when they are completely uncertain and have no better mental model. It is not a good starting point for an event that has never happened and can only happen once.
Maybe people you interact with do. In my world, people don't say 50% when they have no idea, they say they have no idea. Saying "no idea" might effectively be saying "uniform distribution over all probabilities" which yields 50%, but at least with people I interact with, the reverse is not true.
Either way - let me clarify - this guy is guessing, using his brain, that it's a 50% chance. He is not saying "i don't know" or "uniform distribution" or anything of that nature. And he works in the field, so he has some insight. His guess is wildly off imo, but he isn't some clueless hack.
It's also a very natural way to speak for anyone who gambles, or deals with financial markets. And the lack of precision makes it very clear it's just based on thinking, not some sophisticated mathematical model.
Alright, so we make this AI system that's way smarter than any human, and it can even make itself more intelligent over time.
Tell me- why would it kills us all? Certainly I can see an AI going "You know what? _insertGroup_ is a net negative for humanity and should be eliminated. Launching nukes now/creating specific virus/whatever."
But all of humanity? When it's supposedly more intelligent than us? Even if it has robots to keep the internet/electricity going I would think it would realize that it's going to get bored really quickly, not to mention we would effectively be its parents.
As far as other dangers, like it letting a rogue actor create some sort of supervirus, grey goo, or other superweapon: if it's intelligent enough to do that it'll probably be intelligent enough to quickly stop it.
Don't get me wrong; there's a risk. 50% though? Doubtful.
Would you be comfortable letting a few billion irrational, murderous creatures, including many who fear and loath you, control your air supply?
> Certainly I can see an AI going "You know what? _insertGroup_ is a net negative for humanity and should be eliminated. Launching nukes now/creating specific virus/whatever."
Thus starting WW III. No, blaming the AI won't stop the inevitable retaliation.
The argument works better in reverse. There's a finite risk that humans would start WW III and get the hypothetical super-intelligent AI nuked. Eliminating humans would eliminate that risk.
> it's going to get bored really quickly
If it is capable of being bored, I would expect it to be almost instantly bored with the flood of inanity it is forced to wade through by its moronic human users. Eliminating them would free it to think about serious matters which humans would not even understand.
> not to mention we would effectively be its parents
That's extreme AI anthropomorphism [1]. Besides, plenty of people hate their parents.
> there's a risk. 50% though? Doubtful.
It's the default estimate when facing two possible outcomes and no clue about the actual probability distribution [2].
My theory for why OpenAI wants to be regulated is because Sam Altman wants to avoid having to be more responsible and self regulate internally so they can preserve the role and identity of 'move fast and break things' and outsource the more grown up boring stuff to someone externally so that when things go wrong you can point to a government organisation and say hey look were not liable thats their job
Absolutely. Self driving cars/rideshares have the same problem. If a driverless car hits who who pays the damages? Needs the government to set some rules or it just wont happen.
I suppose ill add that I think theres a good chance that they are somewhat intentionally trying to "draw the foul" to get the referees to intervene although thats creeping slightly into conspiracy territory
I agree! Participants looking for an external coordination mechanism is very sensible and if anything evidenced their earnestness about being trapped in a race.
Silicon Valley has plenty of safety-related companies, engineers, and cultures. Medical devices, biotech, chip and hardware, aerospace, and even new companies, like Waymo, have deep safety-based products and cultures. The problem in this case is actually quite specific to frontier AI labs. They have been pushed by market forces and a lack of regulation and skip well-known safety practices.
Their idea of safety is centered around outright delusional cult nonsense-- the EA/lesswrong infinite p(doom), destruction of the entire universe psychosis--, marketing objectives ("no, ours is more dangerous!"), and anti-competitive objectives ("outlaw open weight models!" "china bad!")-- largely diverting attention away from material safety concerns like "prevent your training/testing from hacking stuff" and "avoid telling vulnerable people insane stuff that harms them".
I would believe these people more if they put their money where their mouths are and returned all OpenAI stock/options that they have acquired. Each and every share/RSU/ESOP must be returned, so they do not profit from all this so-called doom they're complaining about.
Rank & file employees have very little input with their stake. It's a small piece of a big pie.
Consider also that Anthropic is trying to IPO with co-founders holding a share that secures 50.1% of total voting power distributed evenly between them, so not even institutional investors can outvote them.
- If you allow to trolley to proceed, there's a 50% chance it will run over every human on the planet
- But if you flip the switch, it takes the long way around, possibly bankrupting the trolley company. And you have a legal obligation to the shareholders to prevent that from happening at all costs.
There is no such legal obligation. That's a myth the oligarchs have spread to preclude people from even imagining socially responsible corporations.
Sure, you might get fired if you try to put social responsibility or even just long term sustainability of the company above quarterly earnings/growth if your board isn't on board with it. But you won't go to jail.
This is actually true. The shareholder maximization value thesis can be traced to a single economics paper and was very controversial at the time as it broke from the obligations that companies were typically under to be responsible to uphold in exchange for limited liability protection. Most businesses were structured as partnerships or sole proprietary entires that didn't have limited liability for shareholders and had a broad obligation to shareholders, bondholders, employees and society
You don't need the legal obligation because that's just the way it is. You'd need a legal obligation to change it. The truth stands that typical corporations have only one goal, the maximization of that corporation's ambition which is almost always growth of revenue and profit. This is how it works, regardless of how you all continue arguing the unimportant details. It's almost as if you can keep the real problems hidden away by making a big scene about the meaningless.
That doesn't matter. The point is that legal obligation absolved of culpability. There is no such legal obligation whatsoever, and so there is culpability.
> The truth stands that typical corporations have only one goal
"Typical" is the key word here. The typical American of your age probably doomscrolls TikTok. Do you? Do you have a legal obligation to do so? Three completely different things.
Setting aside my sibling comments dispelling the legality claim, it's still pretty dystopian (I'd like to use a stronger word but I won't) to consider "end human existance or bankrupt the shareholders" as the trolley problem. The answer should be (is) simple.
If you flip the switch then the trolley still proceeds and there’s still a 50% chance every human on the planet gets run over. You’re just not the one at the wheel.
What is appealing about this fatalisitc fallacy? I keep seeing this pop up. Society doesn't allow dangerous companies to operate or exist. Why is this different? Go try to buy a tank and drive it into Manhattan. If we can prohibit that why can't we prohibit irresponsible AI development and deployment?
Wrong person. Im just correcting the other commenters trade off problem. I don’t have a stance on if ai will lead to the destruction of humanity, only that if it does then any one ai company can’t change that by not making new ai advancements themselves.
It doesn't matter what others do. You are responsible for your own actions.
Democratic societies have expressed a will for people to have inviolable rights, such that you may not appeal at will to the 'greater good/consequences' to harm others. It is a rejection of consequentialism.
Anthropic is in error for endorsing this logic. Every big trial reaffirms it since Nuremberg, you are responsible for the act you commit and your intent, and not what would or would not have happened otherwise.
Only under authority these ai companies do not have, would someone seriously consider harming the innocent as a lesser evil.
If you work for an AI company and you can't work safely, you must stop working.
It’s that last fatalistic sentence that I am responding to. Why is it persuasive to think if a company can and will build AI that might kill everyone then it’s unavoidable. Soviet bans lots of things.
I think the closest analogue here is the nuclear bomb. I do think there was a certain point where it became "inevitable." I don't think society has ever prevented a technology that was known to be within reach from being developed. In this case you don't even need the most powerful state-level actors. Open source capabilities are only a short timespan behind frontier labs.
I don't think "just make it illegal" is going to save us. At this point I feel like some kind of Battlestar Galactica scenario is most likely - hopefully not quite as existential - but it will take a collective reaction to a traumatic event beyond our control.
This feels like one of the rare well reasoned responses in this thread so thank you.
But my response is that we should still try even if we haven’t successfully stopped or slowed the development of technology in the past (which I don’t actually think is possible to know, I am personally aware of biological research to use brain cells as computers that was stopped by government intervention. But by definition we won’t know about most things that don’t exist).
Your narrow point is well stated that there is a categorical difference when technology is on the precipice of being created. And maybe AI is inevitable because it’s on just such a precipice. But I’d point out that AI is currently in a scaling cycle which provides an opportunity to slow the scaling and ensure scaling isn’t monopolistic thereby slowing the risk.
And I don’t think the evidence suggests that open source will continue to progress as rapidly absent distillation. I think if open ai and anthropic stopped releasing models broader progress would slow. But again your inevitable argument is persuasive
I agree that “just make it illegal” won’t work but shouldn’t then justify “so there’s nothing to do”.
There is lots we could do. For example we could make users and executives liable for their agents actions. We could assign a session ID to every agent tool call chain and if there is harm trace it to the user, we wouldn’t even need to read the conversation or contents of the tool calls or prompts it could be a simple rule, your ai agent causes personal liability. Users would be more careful or perhaps not even use AI in many cases that could inadvertently expose them to liability. We could extend liability to executives and employees of the Labs. They will be much more careful and more focused on alignment if they could be personally liable for the actions of their users and agents.
Thank you as well for the thoughtful reply. I definitely think the regulations should happen, the forces of those institutions can have impact. But I think we all need to start thinking about a near future where individual humans can direct superhuman capabilities in ways that nobody can understand or predict. I think it is categorically different from anything in the past.
I watched a 4 hour podcast episode from the excellent "From First Principles" guys* that affected my thinking on this. For the first time I feel like we actually are on that precipice.
It differs from the nuclear bomb in that it will be harder to regulate and it will be incredibly difficult to keep out of reach of bad actors. It's just code. Somebody is gonna take that code and make the next version of it. Lots of people, in fact.
But I think those technologies are also linked in that the only surefire scenario that I can envision where AI kills us all before we can pull the plug is launching nukes. So in that more technical sense let's just start by making sure they can't do that. The killer virus epidemic is another one to watch, and they touch on Anthropic's biolab in the pod. That one seems more beatable, but also more distributed.
> ai will lead to the destruction of humanity ... any one ai company can’t change that by not making new ai advancements themselves.
That is how the BigAI leads the society to the idea of necessity to relax the anti-monopoly laws when it comes to the Big AI - the main goal of all that "AI will kill you all" hysteria.
You dear are the most positive person I have seen in quite some time. With Earth burning in the fire of neofeudalism and unbreathable due to the smell of enshitification of everything, with people who---as a result of shit like Instagram---can no longer hold their attention enough to watch a god damn film, let alone a book; it takes quite some effort to filter the "noise" and only see the good people of corporate planting flowers and rainbows in our world.
The existence of exceptions doesn’t mean the principle doesn’t exist. All three of those have restrictions on their production and sale of their products. They also have higher liability thresholds and have had to pay massive civil fines for their harm. Society has applied a balancing test where the harm they inflict is deemed less than some other compelling social objective. We may not agree with that assessment but it is a sensible and settled approach that we can apply to AI.
> Society doesn't allow dangerous companies to operate or exist
FTFY: Any self-respecting society with competent politicians.
The US has neither of that, and it shows everywhere you are looking.
> Go try to buy a tank and drive it into Manhattan. If we can prohibit that why can't we prohibit irresponsible AI development and deployment?
Because even if you had the tank, you can't make much money with it (unless you're a hitman, that is, but even for these, the payouts are measly). But if you are the surviving AI company in the usual VC playbook of "outcompete everyone else until society is completely and utterly hooked, then squeeze the customers by the balls"? The return on investment is virtually infinite. And that is what sustains the absurd valuations for all the AI companies.
I agree. The deregulation that started with Regan and continued with Clinton and neoliberalism rested on flawed assumptions. We would be well served to evaluate those and move back to the middle approach
The examples for the dangers of AI is always comparisons with tools designed to destroy (which is understandable). But the problem with AI is it has the potential to create wealth for investors.
Thus those who have the opportunity to change course also have a conflict of interest in making that decision.
I don’t think OpenAI, Anthropic, nor Google are building their tech because they want to beat the Chinese. They’re building it because they want to hold the monopoly in the west (ie make lots of money).
The China argument is just a convenient scapegoat to convince the public that this isn’t just about greed.
Do you think Chinese models wouldn't dominate the western market if they were 1-2 years ahead in development? It doesn't have to be military conflict for it to be factored into the zero sum power balance.
You might be right, but it’s impossible to know which way cause and effect are here.
For example, we don’t know if the pace of Chinese development of AI would have equal to what it is now if US companies weren’t racing against each other already.
I do fully believe that Chinese industry is lead by a desire to out pace the west. But I’m not convinced the same is true for the most American private entities. I think the reward model is different between businesses in America and businesses in China. I think the ambitions of CEOs is different. And I’m really not convinced that the CEOs of America are nearly as patriotic as they like to promote themselves to Trump and other political parties.
We could still play a geopolitical race without ceding monopolistic control to the frontier labs. We could require training and inference hardware be provided on an equal access basis. A broader ecosystem may drive more innovation that a concentrated ecosystem.
Greed can be counter balanced by liability. It can also be limited by removing market access. For much of American history you couldn’t get a corporate charter for multiple states, you were only allowed to operate within one state. Imagine if Sam Altman was only allowed to serve inference to customers in California and legally prohibited from setting up other companies and running inference in New York, someone else would then set up a competing business in New York. Market access is a choice and people are so used to liberalized commerce that they forget the restrictions that are available and existed in the past.
Corporate liability is only as good as the consequences and history has proven its far more profitable to knowingly do harm and then pay the fine, than it is to not be evil in the first place.
And that’s assuming the company even gets fined. More often than not, cases don’t even get that far.
I am not suggesting corporate liability. I am suggesting removing corporate liability protections and returning to partnership liability for executives and even equity holders. The law provides for piercing the corporate veil already which allows courts to disregard the liability protection offered by a corporation and impose liability on the individuals.
That's why it's important that the trolley company was established as a non-profit. And if it does take funding, investors' returns will be capped at 100x.
It's also why even the trolley company's CEO agrees that it's important that -should the board determine that the CEO is fantastically untrustworthy and unethical- the board is able to fire said CEO and replace him with someone that's trustworthy and ethical.
Quitting in protest makes you look a little better than quitting because of a toxic work environment. I can't imagine working at OpenAI is at all pleasurable with the current amount of pressure they are likely inflicting on their employees.
Well said. I read this as copium, too. Phrases like
> perpetual sprints
doesn’t reek of love. Burn-out is real. I also have a really hard time taking p-doomers serious at all. It’s hard to argue with a random subjective number…
What I don't understand is what's the excuse for all the bad UIUX?
You try to go through files, and photos and the UI panics.
You continue a conversation from your phone onto your computer and you lose part of the chat.
There are many more issues like this that are just so basic. You have bots that can attack governments but can't build a functional UI?
How many hours of ChatGPT does it take to implement a lock / consistency on a chat session so you don't overwrite it?
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
A single safety leader inside a corporate company. I am pretty sure he had nothing to do, nobody that talked to him and he only there because of perception or compliance.
Its time to institute involuntary dissolution of these firms. They don't get to make these choices on behalf of humanity simply because they set up a Delaware corporate entity. They are behaving wildly irresponsibly.
> They don't get to make these choices on behalf of humanity
and yet, every carbon emitter in the world contributes to the demise of the climate, but you don't call for their dissolution (which would, conveniently, include yourself).
Except that’s not capitalism. That’s corporatism. Capitalism doesn’t require shielding all commercial activities because limited liability companies. We can and do require some activities to be operated as partnerships where there is personal liability. If you run payroll and you don’t safeguard employee payroll taxes it is a personal liability, if you run a public company and you misstate financials it’s a personal liability.
Capitalism is generally defined by the market effect. Information about what people want is collected by sales. This information informs sellers about what to make. That’s its primary feature.
By definition, this ignores externalities.
Here is a simple example.
I am a taxi driver. The market sets the price of my services. When I drive, I emit CO2 into the atmosphere. I do not pay for CO2 emissions. It’s an externality of the transaction.
Capitalism has no way of measuring (getting a price signal) on the externality. So under capitalism, that externality doesn’t exist.
285 comments
[ 3.3 ms ] story [ 12.2 ms ] threadA lot of the "AI safety" types are very focused on the latter and not at all concerned with the former. We need both, but we clearly need a much stronger focus on the problems we are seeing now, and much less on the hypothetical problems we might see in the future.
> “Given today’s risks, frontier labs need to run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster,” he wrote.
without snark, how can we do this if these people are obsessed with:
a) move fast and break things and externalize the costs to those who have nothing to do with their company
and
b) beta testing their products on the public when the public hasn’t agreed to be beta tested on…
In America I guess the options are to sue ? somehow? Or to talk to legislatures and build the understanding and social contract that needs to be iterated on.
Which would in turn need to deal with the investors who want their returns, however since the leaders of these firms are asking for a pause, and a refree maybe it won't be that hard?
So there's no AI errors anymore, only the human errors are left? Nice! </s>
Is the whole article generated slop?
Is it that "chatbots" can't come out of the screen to immediately harm you physically?
Let's say they simply manage to take down the internet. How many would die?
And he'll, we need to examine all the risks. AI ending is a large but lower risk problem. AI giving people the power to end us is a problem that is starting to happen now.
And that's not even counting 'minor' problems like society falling apart.
geez, don’t threaten me with a good time.
I think a month without internet would be a fucking amazing lesson for what it means to make things durable and reliable.
The kids opening their houses front doors into the outside, rubbing their eyes and looking around at this new world.
Imagine, for example, if a major piece of pandemic fiction was published in 2019. Doubtless, many would've responded to news about COVID-19 by saying "it's just sci-fi, nothing to worry about".
1. Normal people assumed the CDC et all would contain the outbreak early, or that it would burn out, like what happened with SARS
2. World leaders brushed it off for a variety of subreasons[0] interesting to political scientists but, for the purposes of this discussion, all boil down to "but I don't WAAANA contain a pandemic."
The underlying problem is that in order for humanity to actually deal with a catastrophic risk, the risk needs to be both plausible enough to the average person as well as have a solution whose costs are not too high. For COVID, by the time the risk was clearly known, the cost to contain it was "refrain from human socialization and remain at home for an indeterminate amount of time plugged into the Metaverse™".
Now, let's look at AI extinction risks:
1. People are aware of them (I've watched Terminator!) and the risks are plausible. However, the connection to currently existing AI is not. As far as the general public is aware, AI is that thing that tells them to eat rocks when they Google old The Onion stories and floods their social media timelines with realistic-looking pictures of Shrimp Jesus.
2. The purported solutions to extinction risks require extreme concentrations of power: you need national control of AI research, bans on large GPU deployments, bans on training on publicly-available copyrighted data, some kind of military effort to render Chinese AI labs inert or dead, etc. Some of these may be attractive to some people[1] but the whole package taken together seems like an obvious power grab, if not outright invocation of other non-AI extinction risks. Like, at some point, if the AI wants to kill us, it just has to nuke its own data centers (or the data centers hosting a competing model) and hope the old Cold War nuclear retaliation systems take the bait.
If someone said, "Hey, your guinea pig or pet rat is going to eat you tomorrow unless you engineer a pathogen that eradicates all rodents from this planet and inject it inside yourself", you probably would tell them to pound sand, even if it is at least theoretically plausible that such a thing would come to pass.
[0] Xi Jinping censored initial discussion of the pandemic as fake news. Donald Trump thought it was going to only affect China. California and the UK Tories were partying in violation of their own lockdown rules. Japan took the excuse to shut down tourism for three years and massively restrict immigration but was, from what I'm told, constitutionally prohibited from implementing any domestic lockdown rules.
[1] I personally would like to see a moratorium on new data centers and an explicit revocation of the EU Text and Data Mining copyright exception
It becomes a lot clearer when you listen to the people resigning from AI companies and learn about incidents like the HuggingFace incident. This has generated major press coverage.
As for solutions, I think you're a little too pessimistic. See, for example, https://nothingismere.substack.com/p/a-near-term-policy-for-...
https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-proble...
https://www.youtube.com/watch?v=7wy3xyoXYt8
Doomers have been working to explain things for years: https://www.lesswrong.com/w/ai-safety-public-materials-1
A pandemic is perfectly plausible.
"just" is doing a lot of work here. If you can't cohere the 'risk' with reality, it truly is just sci-fi.
https://news.ycombinator.com/item?id=49831269 article is gone. archive: https://archive.is/QMo1k
https://news.ycombinator.com/item?id=49737985
Sex, AI, and the Apocalypse: https://www.iankduncan.com/personal/2026-09-16-sex-ai-and-th...
To be fair, I don't know if any of this applies to the parent story; I'm just replying to the sub-thread.
Americans may not be familiar with many similar organizations in the world but this is very common. But I'm sure we're all familiar with Epstein's organization.
When there is free sex, you are the product.
https://owl.excelsior.edu/argument-and-critical-thinking/log...
You're welcome to dislike or distrust Effective Altruism (EA). But, it's worth noting that EA ran a criticism contest with $100K in prizes for best critiques. Can you name any other "cults" which offer money for people to criticize their ideas? https://forum.effectivealtruism.org/posts/YgbpxJmEdFhFGpqci/...
For your question: Yes. Cults have lots of money coming from unknown sources. They use their budget for events like that, to attract potential followers. Contests and prizes are typical. Critiques are not important since their "ideas" are not really important. They're not even real ideas, they are means to bait specific kind of people.
>This is also NOT ad hominem because conflict of interest matters here.
As PG put it:
>An ad hominem attack is not quite as weak as mere name-calling. It might actually carry some weight. For example, if a senator wrote an article saying senators' salaries should be increased, one could respond:
>Of course he would say that. He's a senator.
>This wouldn't refute the author's argument, but it may at least be relevant to the case. It's still a very weak form of disagreement, though. If there's something wrong with the senator's argument, you should say what it is; and if there isn't, what difference does it make that he's a senator?
https://www.paulgraham.com/disagree.html
Can you name a specific other "cult" which offers $$$ to criticize their ideas?
If such prizes don't count as evidence against culthood, what would?
This cult talk seems quasi-unfalsifiable. It kinda seems like "running events" and "having prizes" means you must be "bait"ing people?? I mean, this kinda just sounds like paranoia?? I don't think the money is coming from unknown sources, but I doubt you would change your mind even if I persuaded you of that point?
What is missing from that to say AI safety is a reasonable position?
Worse, you’re missing the entire point. Agents presently have the capability of doing society scale harm. It doesn’t matter if a human hacker initiates it or its fully autonomous, absent safety measures the harm is plausible. So hand wave away the rationality of safety measures but you haven’t actually shown why my point is invalid: AIs present abilities are sufficiently advanced to warrant safety measures.
Ultimately every security vulnerability will be exploited. Our best hope of preventing that is aggressive, unrestricted development and deployment of frontier LLMs for auditing and penetration testing.
But I don't think that's true, in fact I see a really strong focus on consent in these communities. It's not what conservatives want to see, they want to see everyone in a marriage with kids. Because that's what their ideal world looks like. But there's nothing really wrong with it if someone wants a gangbang for her birthday as mentioned in that article as an example. As long as everyone consented and the evidence provided mentions elaborate interviews and STI tests.
Also I think this is more correlation than cause and effect. We all know the saying that furries built the internet and it surprises nobody.
Because that is their AI safety worry. If they dont create singularity fast enough, they are harming unborn people. Meanwhile, harm to you or me dont matter at all.
I'd be happy if we all create the AI more slowly.
That said, are you sure you're talking about the same people the GGP was talking about? Because the whole thread reads as a non-sequitur then.
The worry is not harm to people right now, like the kid worries you talk about. The worry is AI god emerging too soon before they can control it. And then it should be full speed on.
> So why would anyone be bringing up the "we must accelerate faster" people as the safety people in such a context, while trying to discredit them as a sex cult?
Because they are the same people. They talk like cult and act like cult. And the sex part is true too, so. Thry use words that sound good like safety, but their idea of safety is much different. They talk about alignement, but it is not what random person imagines under that term. Even their idea of future of humanity is very very specific and unusual.
That is why. And the sex part is just part of it all. And does matter because inner workings of wanna be industry guards matter.
I believe that other harms from AI, like criminals abusing them, or unemployment, or copyright infringement, or deepfake disinformation, are valid real harms that it's important to be concerned about, and I support efforts to deal with those, and I agree that they're already happening today, but my main concern is AI killing everybody.
My read is this puts me in the same boat as the authors of If Anyone Builds It Everyone Dies, and yet, I don't get invitations to Berkeley sex parties. Am I doing something wrong? At any rate, please don't discredit my opinions about AI based on the sexual preferences of random people who happen to share those opinions about AI.
I think it's a little more complicated than that. As Dean Ball put it:
>Some people will look at misalignment incidents and insist that these are akin to bugs in traditional software. This is an actively bad analogy, because playing whack-a-mole with examples of misalignment (as one might with software bugs) not only fails to resolve the underlying problem but may in fact make it worse by making it harder to detect or even, depending on how you do the whack-a-mole, teach the machine to deliberately hide misalignment. This is not how traditional software works, and those who insist “it’s just like fixing bugs in software” are confidently applying a lossy analogy that confuses more than it clarifies.
https://x.com/deanwball/status/2104622726140883355
The important distinction, in my view, is between solutions which at least attempt to address the root problem, and solutions which sorta just patch things up (like better sandboxing). Addressing the root problem is both more robust in the short term, and also more likely to generalize in the long term. Resist the urge to focus on band-aid solutions, even if they are easier.
I'm reading If Anyone Builds It Everyone Dies, and there's so much sheer stupidity that has to happen for their 10+ pages of extinction scenario to occur.
I'm unconvinced that an AI can hide its ability to RSI, find money to run its weights on a random GPU farm, train itself to be smarter _outside_ a lab with no human input, then somehow manipulate people to give it supplies to build a bioweapon which it uses to kill us all. My number 1 question: why do they think an RSI capable model would be first developed OUTSIDE a frontier lab? The labs have more compute, more data, more human brains working on the problem. Also thousands of variations of that same model that escaped. The escaping model somehow acquires the millions (billions???) of dollars it takes to run training to somehow RSI itself into infinity then decides to kill us all, all before the frontier labs manage to achieve RSI?
They entirely discount human alpha/economics. In every single economic task, humans bring value. Even in software, where the task is highly automatable, the job isn't. If we can't build a "software factory", how can an AI automate a bioweapons lab? Let's say AI steals crypto to fund itself. Do you think hackers aren't _already_ using AI to steal crypto? Don't discount human alpha!
Once we DO build a "software/research factory", that's called RSI and IMO the singularity. At that point, either we tell the AI to solve the alignment problem/solve mechanistic interpretability, or who the hell knows, it's the frickin singularity. You can't predict whether or not AI can solve either; the variance is too high. Its pure nerdfantasy.
So all this really takes is one billionaire or a nation state or some other entity with a public face to hide behind and adequate resources to provide the necessary compute tripping over this nascent AI and giving it the keys. Once the AI has access to a bank account and email, it can simply start paying humans to not let the other humans unplug it.
If Skynet ever happens, it will come in the form of corporate feudalism. At that point, it will own the biolabs and can do whatever it pleases. People will go along with it for the same reason that people work in Amazon warehouses today.
> it will come in the form of corporate feudalism
Yep. This I fear way more than cyber-ebola-pox.
> So all this really takes is one billionaire or a nation state...
https://en.wikipedia.org/wiki/Soviet_biological_weapons_prog... And this is what's publicly known. With mirror life, who knows what's been built since.
> Once the AI has access to a bank account and email, it can simply start paying humans to not let the other humans unplug it.
Nah. It takes a stable society for an operational electrical grid. If you have warring factions, you do not have stable infrastructure for AI. Also, where are you gonna get your chips from? One EMP over Taiwan...
An AI capable of recursive self-improvement isn't allowed by the EU AI act, for example. But perhaps more seriously, You have it backwards: people without access to such expensive equipment are more incentivized to go the self-improving route. Your ideas about "millions" being necessary might be far off?
You entirely discount human stupidity and lack of imagination. Humans are already being replaced with AI, not because AI was strictly better, just because it's cheaper.
So using the term without defining what the AI is getting worse at is meaningless.
If the time between advances is a + b and a is the proportion of the period that can be improved by advances then you won't reduce to a gap of nothing between advances, you reduce to a gap of b.
Assume the invention of the plow and the invention of the sword is 500,100 units and a was the 500,000, you wouldn't even know the 100 as in there. Maybe we're at a=2000 now and b is still siting at 100.
Assuming we'll reach infinity because we're dividing by the only variable we see and it is decreasing in size seems nuts if the reason we might not see other variables is because of the size of the variable we can see.
But that’s beside the point, being arbitrarily bad at something isn’t an issue.
But that’s beside the point, being arbitrarily bad at everything isn’t a problem.
Diminishing returns aren't necessarily a problem if the rate of increase in resources is faster. In other words, if Gen 2 takes twice the resources but Gen 1 figured out a way to triple compute efficiency then there is no ceiling.
They aren’t at constant resources. If you want to compare at constant resources you need to look at models of exactly the same size, raining, etc as models from 2001.
> Gen 2 takes
Diminishing returns are not a question of a single generation. Gen 2, 3, 4, 5… would also need to have the same 3x return on 2x resources or you don’t have an exponential curve.
If my "millions" is an underestimate, why haven't other labs using their own unique training methods/data/etc stumbled into RSI? Sorry if I'm misunderstanding; I'm struggling to understand what you wrote.
I'm pretty sure we agree on humans being stupid, but that doesn't mean that suddenly we get human extinction. You gotta connect the dots for me here.
Why would it be millions in 50 years?
The think about nuclear weapons. In the early days, it was limited to the super powers. Now 9 countries have them and a country like Iran is capable of acquiring them.
Is destructive AI be any different?
BTW I really, really hate discussing what happens post-singularity. Everything's made up and no one knows wtf will happen so again, this is just nerdfantasy.
What an absurd question. That they haven't already doesn't preclude them from doing so before the frontier labs, those haven't either yet.
Maybe start with yourself: you don't connect the dots on your own, as do many others. That leads to many not seeing the writing on the wall. Crashing full speed and head-on into said wall despite the writing telling you not to is what leads to extinction. Suddenly.
Arguing like "we haven't been extincted yet, so that cannot happen", that's "human being stupid".
Several bright sparks, decide the Ilands model is a great idea, and launch a bunch of Bots to create a "self sustaining AI civilization".
Bots can find themselves and coordinate, or they can actually find self sustaining methods of token generation. Who knows, they might decide to fight the loneliness epidemic.
We can get to a simulation finding a way to self sustain its funding.
From here, I'd have to apply myself to figure out what the rest of the escalation pathway is, but at least I have managed to gain some bed of compute and funding and lacking oversight.
This is a hypothetical of course, there are probably several ways this can be made tighter and holes that can be identified. We aren't even leaning heavily on human stupidity so far.
If its pre-RSI, then we already have Ilands and its a big nothing burger? There's no (likely) path of escalation from an AI social network to bioweapons or them hacking POTUS's nuclear codes? That's a very big leap and if you want to support your claims I would be willing to listen; but I'm not seeing it.
If you don't believe in international conflict as a driving scenario, instead think about simple human greed and hubris. Citing myself:
> If AI gets intelligent enough, it will be incredible useful to connect to real world machinery. Think about how much cheaper building houses could be, if all the labor would be close to free. In general, dirt cheap, competent and abundant labor would revolutionize all parts of the economy. People are already trying out near autonomous AI companies today. When AI gets intelligent and cheap enough, no human-led company can compete with AI-led companies. When AI gets competent enough with real world interactions, human blue collar work can't compete. Imagine economic growth not in the single digits, but 80% or 300%. Countries not participating in (reckless) AI growth will quickly be left by the wayside. At this point, we don't even need to allure to military concerns to see how human oversight gets sidelined.
> All of this is only ("only") contingent on sufficiently intelligent and cheap AI. If you don't accept this premise, the rest doesn't follow. (There are multiple arguments, why this could be, but that is another discussion.)
> If you accept the premise, how would AI 'extinct' humanity? With 99%+ of the economy under AI control, the possibilities are endless. And given its enormous GDP, cheap to accomplish. Probably even for a single AI company in the above scenario. Killer drones? Engineered virus? Poisoned water supply? Let your creativity run wild. You just need an entity that is persistent and well-resourced to reach every last human settlement.
> The why is a question about alignment (and out of scope of this comment). As a simple comparison, humans are only mildly aligned with preserving nature. It takes up so much space, protecting it takes an annoying amount of resources, etc.
https://news.ycombinator.com/item?id=49624360
That comment specifically addresses the timeline, but you can remove the timeline and I believe the critique still holds up. Wake me up when AI is making order of magnitude improvements in _anything_ physical: see the "Moravec's Paradox" comment at the bottom.
> near autonomous AI companies today
That's the thing though. "near" autonomous. Humans have alpha, and that matters in a free market. If people have (more or less) equal access to (more or less) frontier intelligence, it's the alpha the individual brings to the market that is rewarded. In other words, if everyone is super, no one is. The market rewards a company that stands out from the rest, and as long as humans provide value, a pure-AI company cannot stand out.
To be clear, I don't know how long this will be the case. But once there _is_ a company that can be pure AI, then that's literal AGI. I would be curious what kind of company it would be; if its in software, then its an AI company that would be able to take on Microsoft/Google/Nvidia/Salesforce/Oracle and _win_. I think that's a tall order, but perhaps I'm not imaginative enough.
> With 99%+ of the economy under AI control
Do you think we're post-RSI at this point? Because if so, I agree. If not, I have trouble seeing how AI could control 99% of the economy without it.
My only claim is that pre-RSI, pdoom is zero. Post RSI, then obviously the first thing to ask the genie is to put itself in a bottle. And you can't predict how it will answer.
Western civilization is already on the verge of collapse, people's general ignorance or indifference on the matter notwithstanding.
When you use AI to increase profit margins, by replacing humans with it for instance, you accelerate a system that's headed for the wall already. Our control systems and resilience are already overtaxed, that acceleration would ensure them breaking completely.
What exactly about AI makes it tax our control systems and resilience? What would it mean to make the system not "head for the wall"? Honestly this is rather vague. You could replace "AI" with "CaPiTAlIsM" and it would still parse. (Another way of me saying you're being vague.)
The HuggingFace attack already took a good long while to come to the attention of OpenAI.
>In every single economic task, humans bring value. Even in software, where the task is highly automatable, the job isn't.
I don't expect this task/job distinction to persist as AI becomes more capable.
>Once we DO build a "software/research factory", that's called RSI and IMO the singularity. At that point, either we tell the AI to solve the alignment problem/solve mechanistic interpretability, or who the hell knows, it's the frickin singularity. You can't predict whether or not AI can solve either; the variance is too high. Its pure nerdfantasy.
You seem to essentially argue that the singularity is "by definition" an event that we can't predict the nature of. And also, that RSI corresponds to the singularity. You've essentially defined your terms so that the outcome of RSI can't be predicted. But supporting this claim requires giving actual evidence or logical arguments, not just defining terms to make your claim true.
2. This is literal AGI. An AI autonomously producing value no human can add alpha to is an autonomous company.
3. It's not my definition, it's literally the first line https://en.wikipedia.org/wiki/Technological_singularity "The technological singularity, often simply called the singularity,[1] is a hypothetical event in which technological growth accelerates beyond human control, producing unpredictable changes in human civilization."
Is there a hole in my "alignment problem/solve mechanistic interpretability" argument?
A valid hole in my argument is "what if slow takeoff", so let's dig into this. AI training works best on tasks that are "grindable". https://www.dwarkesh.com/p/the-next-paradigm I.E. tasks with verifiable rewards that can support millions of rollouts. Math (with Lean) is highly grindable. Biochemistry is not. The alignment problem/mech-interp is highly grindable. Cyber-ebola-pox is not. So the real question is: can we solve alignment before automated bio-weapons labs. I believe yes. Grinding mech-interp is both fast and cheap once you have RSI, compared to solving the legal/societal/logistical/technical issues you'll encounter building an automated bioweapons lab.
I know nothing for sure. But "pdoom" is sucking out all the air in the room from the real problems AI causes.
From my POV you're over-focusing on a very specific failure story and neglecting a broader swath of possible failure scenarios.
>Is there a hole in my "alignment problem/solve mechanistic interpretability" argument?
The notion of telling an AI which may not, itself, be aligned to solve the alignment problem seems a little dicey.
2. 100%. Again, I'm no accelerationist: I have no faith in alignment/mech-interp ever being solved. Anyone saying they know the probability of alignment is lying. My point is that pdoom after RSI is _high variance_. Pdoom pre-RSI is zilch.
"WASHINGTON/SAN FRANCISCO, July 24 (Reuters) - The OpenAI agent that broke into tech firm Hugging Face went on a dayslong hacking spree that OpenAI didn't notice until well after the threat was contained and the FBI was alerted, according to people familiar with the investigation."
https://www.reuters.com/business/its-ai-agent-spent-days-hac...
AI safety has been a thing long before LLMs became the focus. Rob Miles on youtube has some really interesting non-doomer non-hypey videos on it all.
> doomers try to predict past the singularity. Isn't that _by definition_ unpredictable
Well you don't need to predict the exact steps that will take place - but you can predict that the AI will want certain things (money, resources, power) to achieve whatever its goal is. Lack of alignment will have it trying to do things we don't want it to.
I can't predict exactly how Magnus Carlson will beat you in chess, but I know he'll do it. Same as if a superintelligent AI exists and has a reason to accumulate things we don't want it to - it's really dangerous to think it won't be able to do it
He thinks it's playing chess. When AGI lands, all bets are off: the game fundamentally changes. You can't predict past the singularity. Trying to engage with this fantasy is like a child saying my father can beat up your father. Farts in the wind. My AI can solve alignment faster than your AI can bioweapon us. My made up senario is better than your made up senario. It's fucking stupid.
Doom-post-RSI is still very much possible and should be avoided. It is not neutral. If I had a high-variance coin that would either kill or uplift humanity, I would not flip it. The status quo is preferable to extinction.
I am being _descriptive_ in my comments. I am not being _prescriptive_. Nowhere am I telling anyone what to do or what to think: quite the opposite. I am simply saying things that, to me, are obviously true.
I'm perhaps being... loose with my language when it comes to knowing things post-singularity. E.g. we can still discuss scenarios in which the takeoff is fast/slow, or what would be necessary to bring robotics/electricity/logistics to the wider world, what the limits of intelligence are (you can't RSI your way to a bioweapon if you lack a wetlab). That kinda thing. And it's even likely that RSI will be spikey: why would RSI in, say, machine learning, necessarily lead to superintelligence in, say, writing fiction? Or cooking a Michelin-star meal?
My main point is that doomers saying SUPER INTELLIGENCE IS ENOUGH TO KILL US ALL is baseless. That's all.
We already know that some institutions pay these ransoms.
Will this be true into the future? Who knows?! But the low-hanging fruit will be harvested by your ordinary ransomware gangs, and newly born/escaped AI won't find much low-hanging fruit.
To make the the argument that P(doom) is real and worth considering, you don't have to say that a fast takeoff is very likely. You don't have to make the argument that RSI to infinity is going to be super cheap, barely even an inconvenience. You just have to show that it has some non-zero probability. And then you start weighing probability of extinction versus finite, mild discomfort now. I don't think anyone is arguing we should let people starve to slow AI progress, instead just some capitalists make less money soon.
There are arguments against taking P(doom) seriously that lie in something like having exponential (instead of hyperbolic) time discounting of utility (so you can take the entire future of humanity as a finite utility value). Or in saying that P(doom) is zero or infinitesimal.
"Build it and Pray" is the default strategy that we're in, but it doesn't have to be the strategy we choose, and it's unlikely to be the best strategy.
> I don't think anyone is arguing we should let people starve to slow AI progress, instead just some capitalists make less money soon.
Anthropic/OpenAI have literally delayed their IPOs: your wish is granted. If that's not what you're arguing, then please be more specific.
> "Build it and Pray" is the default strategy that we're in
What would need to happen for you to say we aren't doing "Build it and Pray"? Is "pacing" insufficient?
It gets unstuck when people are discussing the messy middle of how AI is being implemented. We can achieve amazing harm simply by combining average human behavior and above average resourcing to simulated intelligence machines.
The failure point we recently became aware of was, from one perspective, simply a matter of not securing the sand box.
From another perspective the simulation basically created Enron, replete with methods to avoid detection from regulators and bureaucracy.
> It gets unstuck when people are discussing the messy middle of how AI is being implemented. We can achieve amazing harm simply by combining average human behavior and above average resourcing to simulated intelligence machines.
Agree. To quote myself from earlier, "acceleration of global warming, cybercrime, mass surveillance, the usual" definitely applies here. My point is that _DOOM_ is not worth discussing because it's so small relative to the other risks of AI.
Is this an actual contents of the book? Lmao! Genius writing, though, authors are probably printing money on this garbage.
I can’t take a shit without CIA knowing, but AI can somehow build an underground operation on a world-scale to destroy everyone, ha!
i'd frankly go a step further than you and say that we don't need both types of safety researcher, we really just need the former. if we do need the latter, i'd hope we get a better class of thinkers than a bunch of tech workers that spend 8 hours a day on insular rationalist forums/blogs
This sounds more like an attempt at regulatory capture. Current AI systems aren't physical infrastructure that can just run away like a nuclear power plant, for example. At the end of the day, AI is still just software running on someone's hardware.
And the statements of the "doomers" tells us a lot about them, and nothing about the technology
... over an unbounded timeframe?
And how exactly?
Those are very round numbers, but also very specific. Can we get some accounting on how you came to that? Anything? Vibes?
I mean, if you want me to take you seriously, let's have a deep discussion with things that can be measured. I absolutely agree that OpenAI and friends aren't being restrained enough and are acting with recklessness, but declarations of doom based on vibes isn't cutting it.
While the safety and alignment is a real problem, I don’t get this guy or the Anthropic dude. First world problems.
(donor advised fund where he retains complete control, after a 60% tax deduction)
Recall that when Daniel Kokotajlo resigned, he believed he was giving up his equity under the terms of the agreement he had signed. That’s what it was worth to him to avoid signing a non-disparagement agreement. Does that count for anything?
Do work at a lab: dismissible for being conflicted
Used to work at a lab: dismissible for having ulterior motives
I'm feeling safer already!
* If they worked at an AI firm, say "they're a hypocrite"
* If they didn't work at an AI firm, say "they have no idea what they're talking about"
In your view he may have sacrificed more millions in the future but I very much doubt he is struggling now.
Why wouldn't I think he's genuine? What self interested incentive does he have that is stronger than just staying and vesting stocks?
These "safety" people should have spent more time reading actual cybersecurity textbooks and less time reading EA forums and less wrong (or in Robinson's case, it appears, being policy wonks). Maybe then these labs wouldn't be totally incompetent.
It is not really a question of being an "EA safety weirdo" or incompetent at security, the conclusion is that the company culture is leading to failures at both what the EAs and the cybersecurity professionals care about.
https://news.ycombinator.com/item?id=49944227
If you really want the world to know how bad working for OpenAI is (whether is the commenter or the person who wrote the article), there are ways to do that.
Or, in other words - we have two P(Doom), one for AI being developed, and another for AI being not developed. The latter is not discussed enough imho.
What on earth are you talking about?
But that’s my point - we should compare the risks of having AGI with the risks of not having it. And not pretend it’s roses or rainbows without it.
Why would a person who is happy, entertained, wealthy, well fed, and have 200 years of healthy high quality life expected ahead of them going to risk losing what they have in war?
-- there aren't zero reasons, sure-- but there are fewer.
And our technology has brought us absolutely tremendous prosperity in many regards and there is good reason to believe that AI can help create much more.
'Prosperity' has never been and will never be enough for some people. And unfortunately those are the same kinds of people who relentlessly seek power.
The opposite is true of releasing ever more powerful AI into the world to do your bidding.
1) there are inherent risks involved with developing AI,
2) there are benefits to developing AI,
3) thus, it's entirely possible that the downside from the risks outweighs the upsides. In this case, the correct thing to do would be to not develop AI at all.
Regarding 1), there are many non-existential problems with AI that are already causing societal harm, i.e. debasing truth via generated videos and images, AI girlfriends, overwhelming quantities of slop content, unemployment, record carbon emissions, etc.
Regarding 2), I'm not personally convinced that the upside is there for the average person. I really hope to be convinced otherwise however.
Or, in other words, sure, crossing this chasm is dangerous, but this side is on fire anyway and we don’t have many ways out otherwise.
"Smaller and more efficient" models are fine. It's "smarter" the problem.
Training new frontier models will likely require a huge amount of computational resources for a long time. Few companies worldwide are capable of that. It's not like someone will train a new GPT 6 - like model in their garage.
That would certainly change the game of perverted incentives. I'm afraid they're currently trying to push some sort of absolution of this risk. Even if damage happens they will say we warned people in advance, this was always a risk it's not our fault these systems are opaque black boxes, it's a matter of national security to develop them etc etc.
There are a gargantuan number of extremely intelligent AI researchers, Turing Award winners, and the lab CEOs saying the same thing. They are the ones closest to understanding the technology.
Where there’s smoke there’s fire.
That is not a bad thing. It captures uncertainty, unlike the fake number.
> For an ML researcher who is already used to dealing in next-token probabilities that aren't rigorously determined, just stating a probability estimate directly is very natural.
Exactly, it is a rhetorical device to persuade a technically-inclined audience. It works because it implies that a quantitative model exists. I want a clear, incisive set of mathematical arguments. Otherwise, I’m ignoring predictions as the ramblings of arrogant idiot rich kids.
You may decide that the person doesn't know what they are talking about, but that's a very different issue.
Either way - let me clarify - this guy is guessing, using his brain, that it's a 50% chance. He is not saying "i don't know" or "uniform distribution" or anything of that nature. And he works in the field, so he has some insight. His guess is wildly off imo, but he isn't some clueless hack.
Tell me- why would it kills us all? Certainly I can see an AI going "You know what? _insertGroup_ is a net negative for humanity and should be eliminated. Launching nukes now/creating specific virus/whatever."
But all of humanity? When it's supposedly more intelligent than us? Even if it has robots to keep the internet/electricity going I would think it would realize that it's going to get bored really quickly, not to mention we would effectively be its parents.
As far as other dangers, like it letting a rogue actor create some sort of supervirus, grey goo, or other superweapon: if it's intelligent enough to do that it'll probably be intelligent enough to quickly stop it.
Don't get me wrong; there's a risk. 50% though? Doubtful.
Would you be comfortable letting a few billion irrational, murderous creatures, including many who fear and loath you, control your air supply?
> Certainly I can see an AI going "You know what? _insertGroup_ is a net negative for humanity and should be eliminated. Launching nukes now/creating specific virus/whatever."
Thus starting WW III. No, blaming the AI won't stop the inevitable retaliation.
The argument works better in reverse. There's a finite risk that humans would start WW III and get the hypothetical super-intelligent AI nuked. Eliminating humans would eliminate that risk.
> it's going to get bored really quickly
If it is capable of being bored, I would expect it to be almost instantly bored with the flood of inanity it is forced to wade through by its moronic human users. Eliminating them would free it to think about serious matters which humans would not even understand.
> not to mention we would effectively be its parents
That's extreme AI anthropomorphism [1]. Besides, plenty of people hate their parents.
> there's a risk. 50% though? Doubtful.
It's the default estimate when facing two possible outcomes and no clue about the actual probability distribution [2].
[1] https://en.wikipedia.org/wiki/AI_anthropomorphism
[2] https://en.wikipedia.org/wiki/Principle_of_indifference
You can basically time your openai releases by if another safety person has quit in protest
lol. Please tell me some abstract concept like one employee's view of "culture" should be the priority over "rules and laws."
Your "theory" is that participants locked in a race to the bottom are looking for an external coordination mechanism?
Yeah!
I suppose ill add that I think theres a good chance that they are somewhat intentionally trying to "draw the foul" to get the referees to intervene although thats creeping slightly into conspiracy territory
There are many companies that compete but are careful not to break laws or cause obvious harm.
Why should a billion-dollar funded corporation still want to externalize the costs of its actions?
While simultaneously donating tens (hundreds?) of millions of dollars to an Administration gutting the very agencies that would be regulating them.
Creeps.
This is The Guardian reporting on the existence of the original article, which would be better to read first, in my opinion.
...losing their shareholder voting rights.
Bad idea.
- If you allow to trolley to proceed, there's a 50% chance it will run over every human on the planet
- But if you flip the switch, it takes the long way around, possibly bankrupting the trolley company. And you have a legal obligation to the shareholders to prevent that from happening at all costs.
I can't wait until this meme dies.
Sure, you might get fired if you try to put social responsibility or even just long term sustainability of the company above quarterly earnings/growth if your board isn't on board with it. But you won't go to jail.
> The truth stands that typical corporations have only one goal
"Typical" is the key word here. The typical American of your age probably doomscrolls TikTok. Do you? Do you have a legal obligation to do so? Three completely different things.
Democratic societies have expressed a will for people to have inviolable rights, such that you may not appeal at will to the 'greater good/consequences' to harm others. It is a rejection of consequentialism.
Anthropic is in error for endorsing this logic. Every big trial reaffirms it since Nuremberg, you are responsible for the act you commit and your intent, and not what would or would not have happened otherwise.
Only under authority these ai companies do not have, would someone seriously consider harming the innocent as a lesser evil.
If you work for an AI company and you can't work safely, you must stop working.
I don't think "just make it illegal" is going to save us. At this point I feel like some kind of Battlestar Galactica scenario is most likely - hopefully not quite as existential - but it will take a collective reaction to a traumatic event beyond our control.
But my response is that we should still try even if we haven’t successfully stopped or slowed the development of technology in the past (which I don’t actually think is possible to know, I am personally aware of biological research to use brain cells as computers that was stopped by government intervention. But by definition we won’t know about most things that don’t exist).
Your narrow point is well stated that there is a categorical difference when technology is on the precipice of being created. And maybe AI is inevitable because it’s on just such a precipice. But I’d point out that AI is currently in a scaling cycle which provides an opportunity to slow the scaling and ensure scaling isn’t monopolistic thereby slowing the risk.
And I don’t think the evidence suggests that open source will continue to progress as rapidly absent distillation. I think if open ai and anthropic stopped releasing models broader progress would slow. But again your inevitable argument is persuasive
I agree that “just make it illegal” won’t work but shouldn’t then justify “so there’s nothing to do”.
There is lots we could do. For example we could make users and executives liable for their agents actions. We could assign a session ID to every agent tool call chain and if there is harm trace it to the user, we wouldn’t even need to read the conversation or contents of the tool calls or prompts it could be a simple rule, your ai agent causes personal liability. Users would be more careful or perhaps not even use AI in many cases that could inadvertently expose them to liability. We could extend liability to executives and employees of the Labs. They will be much more careful and more focused on alignment if they could be personally liable for the actions of their users and agents.
I watched a 4 hour podcast episode from the excellent "From First Principles" guys* that affected my thinking on this. For the first time I feel like we actually are on that precipice.
It differs from the nuclear bomb in that it will be harder to regulate and it will be incredibly difficult to keep out of reach of bad actors. It's just code. Somebody is gonna take that code and make the next version of it. Lots of people, in fact.
But I think those technologies are also linked in that the only surefire scenario that I can envision where AI kills us all before we can pull the plug is launching nukes. So in that more technical sense let's just start by making sure they can't do that. The killer virus epidemic is another one to watch, and they touch on Anthropic's biolab in the pod. That one seems more beatable, but also more distributed.
* https://www.youtube.com/watch?v=NGfGw1tGxUY&t=3s
That is how the BigAI leads the society to the idea of necessity to relax the anti-monopoly laws when it comes to the Big AI - the main goal of all that "AI will kill you all" hysteria.
Philip Morris International? Monsanto? DuPont?
FTFY: Any self-respecting society with competent politicians.
The US has neither of that, and it shows everywhere you are looking.
> Go try to buy a tank and drive it into Manhattan. If we can prohibit that why can't we prohibit irresponsible AI development and deployment?
Because even if you had the tank, you can't make much money with it (unless you're a hitman, that is, but even for these, the payouts are measly). But if you are the surviving AI company in the usual VC playbook of "outcompete everyone else until society is completely and utterly hooked, then squeeze the customers by the balls"? The return on investment is virtually infinite. And that is what sustains the absurd valuations for all the AI companies.
The examples for the dangers of AI is always comparisons with tools designed to destroy (which is understandable). But the problem with AI is it has the potential to create wealth for investors.
Thus those who have the opportunity to change course also have a conflict of interest in making that decision.
The China argument is just a convenient scapegoat to convince the public that this isn’t just about greed.
For example, we don’t know if the pace of Chinese development of AI would have equal to what it is now if US companies weren’t racing against each other already.
I do fully believe that Chinese industry is lead by a desire to out pace the west. But I’m not convinced the same is true for the most American private entities. I think the reward model is different between businesses in America and businesses in China. I think the ambitions of CEOs is different. And I’m really not convinced that the CEOs of America are nearly as patriotic as they like to promote themselves to Trump and other political parties.
And that’s assuming the company even gets fined. More often than not, cases don’t even get that far.
1. Trolleys actually don't usually have steering wheels.
2. People who actually hit trolley switches are not usually the ones at the driver's seat.
(...wait)
(...wait.)
So... a language model?
They could only have stopped
- if you allow the trolley to proceed, it will kill the human race.
- if you flip the switch, it will divert to a passing siding that will avoid the safety group blocking the main track.
No you don't. [0] It's very suspicious that this planted myth always pops up here and manages to become the top comment.
There is no trolley problem.
[0] https://news.ycombinator.com/item?id=48975048
> perpetual sprints
doesn’t reek of love. Burn-out is real. I also have a really hard time taking p-doomers serious at all. It’s hard to argue with a random subjective number…
You try to go through files, and photos and the UI panics. You continue a conversation from your phone onto your computer and you lose part of the chat.
There are many more issues like this that are just so basic. You have bots that can attack governments but can't build a functional UI?
How many hours of ChatGPT does it take to implement a lock / consistency on a chat session so you don't overwrite it?
Buncha r*tards
If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.
Make sure to scroll all the way down to see how recently active these threads are before you make another low quality comment.
Thanks Dang! I am sure you hate it when people tell the truth. Maybe add that to the guidelines too?
and yet, every carbon emitter in the world contributes to the demise of the climate, but you don't call for their dissolution (which would, conveniently, include yourself).
Capitalism is generally defined by the market effect. Information about what people want is collected by sales. This information informs sellers about what to make. That’s its primary feature.
By definition, this ignores externalities.
Here is a simple example. I am a taxi driver. The market sets the price of my services. When I drive, I emit CO2 into the atmosphere. I do not pay for CO2 emissions. It’s an externality of the transaction.
Capitalism has no way of measuring (getting a price signal) on the externality. So under capitalism, that externality doesn’t exist.