Humanity isn't ready for the coming intelligence explosion
economist.com
economist.com
I'm not afraid of rouge agents nearly as much as I'm afraid of us building a permanent underclass of people dependent on whatever scraps the people who devalued their labor decide the grace them with. I'm afraid of the security apparatus that will be built to keep this underclass in line.
I think the message the team lead takes away right now is "oh, this can be done a lot faster, let's do more!" This is exciting for a bit, but I think it comes at the cost of learning -- and being able to communicate --architectural lessons that would stage off systemic issues later.
I have been able to build applications that I couldn't dream of before, not because of a lack of capability on my part, but because of time: I don't have the time to learn all of the new APIs and frameworks to accomplish my goal. Now I can build far more complex useful applications, with functions and features that would take me months or more to learn how to code.
The downside is that I didn't learn how to use those APIs, but it can be argued that these skills are low value to begin with. APIs change, there are competing APIs that do the same thing, and you may only use that skill for one project and then move on. Gaining those skills can be a big time sink and may not result in long-term gain.
Think of all the time spent learning the ins and outs of come-and-gone APIs and frameworks over the decades (I chose not to provide examples to save the resulting side-arguments, but I know you can rattle some off in your head). Those APIs and frameworks were bridges to make developing easier for the common developer, but they change so quickly that I'd argue they waste our time, and lead to almost endless re-work as we re-architect our software for the framework de jour. And they grow in complexity over time, ultimately becoming the next bloated framework. It's an endless, wasteful cycle.
Maybe AI can help us get out of that.
Verifying is harder than building!
That's pretty different from "didn't have the time/money/motivation to"
If you didn't have the skills to build something before then you don't have the skills to verify something is correct after
Clicking around a finished UI is not verifying
Maybe start at the first message here: https://news.ycombinator.com/item?id=48553444
I still don't have the skill/knowledge to build cross-platform native applications the way I'm doing it with an agent now. I am still able to verify it's correct. How is that possible if what you say is correct?
> Clicking around a finished UI is not verifying
Definitively a part of it, if you care about having mouse interactions and having graphical UIs.
GP said "couldn't dream of building". To make an analogy, I wouldn't dream of building a house if I had to do every bit of brick-laying myself. But with others doing what I can't do, I can indeed dream of building a house and I do know how to verify that the house was built to satisfy my dream.
This is exactly the naive gap that is happening with AI. There is no trusted middleman who does the verification and takes the burden of accountability. The AI vendor tries to sell it like magic but with a disclaimer that is is merely entertainment. The user tries to pretend they are reviewing, but often is just implicitly trusting the tool to do right by them on things they have no real capacity to evaluate.
It's a great example of a project that needs accountability, but there's also thousands of other projects that need absolutely no accountability.
It's all relative to the stakes. The lower the stakes the less informed verification is needed.
The real trick is being informed enough on the boundaries to know where it matters. In the construction analogy, you need to be informed enough to know that a house is a bad idea since there are safety concerns. However building something small and non-load-bearing is probably fine to "vibe".
Eg there's no expertise needed to judge a garden trellis or 2 foot picket fence. It either works or it doesn't, and if it fails down the road there's no harm.
This boundary knowledge is the important bit imo.
What I’ve learned is quite minimal due to the slop nature. However I’d have never even started this project in the past due to time, so it’s still fun being able to learn small bits of projects I’d otherwise not even touch.
We have plenty of them that humans even find to be pleasant to solve, such as Minesweeper :)
For a generalist, it's a huge force multiplier.
In the future, there will be no more generalists with your capabilities. At all. No one will ever develop the skill. We're at the end of our career's lifetime. We're like the horse watching more and more cars appearing.
Within a generation, the coding languages we use today will all vanish, to be replaced with direct to metal methods, assembly, etc. The only reason C exists, is for humans. PHP, rust, python is for humans. There will be no more compilers. No more human-readable languages. No more scripting with readable code.
No more open or closed source.
Imagine taking every aspect of computing which is designed for humans to understand, and removing it, because it's redundant.
Now once removed, and once humans no longer understand on their own, imagine trying to build something independently? Imagine trying to bootstrap modern compute, without all the layers which came before?
My point is that soon, perhaps 20 years, most humans will be incapable of coding, developing, or working with compute in any way outside of an end-user role. Even those previously skilled, will now be looking at an entire kernel written in assembly.
And that's before new CPUs appear, with new assembly. New assembly, once again, not designed by or for humans to understand. And yes, current assembly is designed that way.
We are not simply looking at a bit of help here. We're looking at the entire end of every single computing job, ever. We're looking at the complete loss of control, of something which is the underpinning of our entire society. Every aspect of everything we have, runs on a chip now. Personal transport, everything in health care, everything in resource extraction, everything harvesting food through to delivery to table, all of it involves some aspect of compute.
And we soon will not comprehend nor create that underlying aspect of society.
I dunno. It all seems like the answer to the fermi paradox.
If they truly can always recognize problems and communicate them in understandable ways, then people's skills won't atrophy.
Everyone's skills will be enhanced.
New developers will learn all of the things they need to learn through using the tool.
It takes longer to build than destroy. Not just trust. Anything. The lack of trust has created friction for everything; more regulations, deeper background checks, purchasing goods and services… the list goes on. Overall, AI is making this situation worse, not better.
That said, I have seen a revolution coming for decades, and I think this AI-paved road to dystopian hell should at least be entertaining.
[0] https://apnews.com/live/iran-war-israel-trump-lebanon-march-...
I agree with you those dangers exist, but you seem to have ignored the specific threat described in the article, from a superintelligence capable of manipulating human beings indetectably, where we'll be at its mercy and only able to hope it's fully aligned with humanity's interests. The risks you describe are grave, though - especially economically, if governments don't grow to outmuscle the power of the oligarchal owners of these systems, which seems wholly unlikely at the moment in the West, with its neoliberal orthodoxies.
And its asymmetric warfare: bad people can destroy in very little time, good people cannot create at the same rate.
Entropy is always ignored in teh battle of "good" vs "evil": you need to defend every day, evil just needs to succeed once.
He searched his pockets. No more coins; nothing.
“I’ll pay you tomorrow,” he told the door.
Again he tried the knob. Again it remained locked tight.
“What I pay you,” he informed it, “is in the nature of a gratuity; I don’t have to pay you.”
“I think otherwise,” the door said. “Look in the purchase contract you signed when you bought this conapt.”
In his desk drawer he found the contract; since signing it he had found it necessary to refer to the document many times. Sure enough; payment to his door for opening and shutting constituted a mandatory fee. Not a tip.
“You discover I’m right,” the door said. It sounded smug.
From the drawer beside the sink Joe Chip got a stainless steel knife; with it he began systematically to unscrew the bolt assembly of his apt’s money-gulping door.
“I’ll sue you,” the door said as the first screw fell out.
Joe Chip said, “I’ve never been sued by a door. But I guess I can live through it.”
- Ubik
Unattended ChatGPT will rarely burn your breakfast or start fires.
More than once I have gone to my toaster in the morning and found cold bread there that someone toasted the day before and forgot about. Because obviously you push the toast down and go do something else. You don’t need to be staring at it continuously, nor are other people impacted by your toast unless they are waiting to make their own. Toasting bread doesn’t have to affect an entire household.
It is not.
E.g. all "atoms" of a glider in Conway's Life get replaced every couple turns but an observer can tell it is still a glider because it keeps the shape, and it the same glider because it continues it's previous state.
This makes AI not quite alive because it's missing the continuity.
In a tortured semantics sense sure, but in a practical sense they refer to different types of things.
Inference engines are trying to converge any input to a certain limit, and I presume that's close enough to an equilibrium to not count as living thing.
I wouldn't like to debate a process handling the SIGTERM signal, or a container engine applying some restart policy.
That was an Ad.
An Ad in the same style of many before it. Not through an ad agency, but the AI companies hyping their product by making their product seem scary and dangerous.
[0]: https://en.wikipedia.org/wiki/The_medium_is_the_message
IE a Ram truck won't make you a manly cowboy with just the purchase.
And the thing that matters is: capability.
Even today's AIs are capable of autonomous goal-oriented agentic behavior - and growing more and more so with every release.
At sufficient capability, AI stops being "a technology we use", and becomes a force in itself - not unlike humankind. Because intelligence is very powerful - it's what allows humans to dominate the world. A world where human intelligence has a peer is a world where human control is contested. And beyond that?
You might end up being more outclassed by AI than a toaster is outclassed by you.
Today's systems are mostly good at following instructions. But push them far enough, and you already get weirdness like alignment faking, instrumental self-preservation and more. We know because we've seen it in lab settings, while probing for extreme failure modes on purpose - but the world is large and strange enough that edge cases like that are liable to surface naturally.
The reason why none of this has exploded in our faces isn't that today's AIs never want to do weird and dangerous things. We know they do, at times. It's that today's AIs are incapable of pulling them off when they try.
Capability is the thing that matters, at the day's end.
There is to the best of my knowledge no fundamental limitation to having an agent/claw go on like this for 80 years with a prompt like "live your life to the fullest".
- Prompted as in prompt made of tokens -> for LLMs, tokens double as a clock signal. Time only flows when tokens are pushed through them.
- Prompted as in specific request placed in the stream of tokens -> Yeah, they do that all the time whether it's getting into infinite loops of repeating same pattern, or suddenly deciding to do things based on inputs they normally ignored.
Also don't forget that everything is a "prompt" for LLM. All input tokens end up in the same place.
In the current UIs is there a lot of suppression then as I have not seen things start on their own?
I meant an LLM doing something without any external prompt at all. Not doing something different etc but rather do something without a token/prompt ever flowing to it.
We shouldn’t be doing that as a civilization in case it works.
Today's AI harnesses just keep feeding AI tokens - and can, in principle, do so indefinitely. "Streaming LLM" is not mainstream but not unknown either. "Automatic context compaction" is somewhat similar in what it does, and very common. "Keeping the lights on" isn't hard. And you have to turn on those lights to have an AI do anything at all.
At the same time: that type of harness often provides a "tool call" interface. Plenty of tools that end up attached to LLMs can do things like run arbitrary code, spawn instances of the same AI, and more.
That's frankly enough. An AI that can spawn more copies of itself can, in principle, just keep itself running indefinitely - even in a non-streaming no compaction harness. An AI that can run arbitrary code on a system can, in principle, do anything a human using that system could do.
The thing that truly limits what an AI can do is the AI itself. Practical AI safety relies on AI being either too weak or too well behaved to cause major issues.
Alive is always relevant. Our moral consideration must place beings that are alive or processes that support the life on this planet above things which are not alive.
In a just world, it must be adapted to us. It’s merely a file sitting on disk with numbers in it. And to put a fine point on it: the ambitions of the wealthy people who make it must be adapted to us.
The only system of morality that applies to LLMs is their effect on us and the planet. Don’t anthropomorphize a matrix.
I don't think that is universal true. Personally I my ethics revolve around how to preserve and spread life in general. I happen to be human and we are the most capable and sentient species so that means creating and maintaining conditions specifically for humans, but if another species comes that is better suited and acts better then it would be selfish to insists on us.
Having said that, AI to me is a tool, not a species and I don't see how it can be ever be sentient. But I also don't know how consciousness works. So I am open for the questions of the future in the future - and now rather focus on the practical implications that AI will have on our current society.
How do they decide between using 'a' or 'an'?
"There is an animal very similar to a crocodile but I cannot remember it's name"
and the model responds with
"I believe the animal you are thinking of might be " ("a" / "an")
Are you saying that it would pick the result fairly randomly and then based on it's choice pick an animal that starts with a consonant or a vowel?
And with what we know of LLMs? Autoregressive transformers are Turing complete in theory, and we are yet to find anything that LLMs are "fundamentally incapable" of in practice. Even continuous learning is already approximated with in-context learning - both allow a system to learn from prior experience, both have practical limits on how far they go. That's what powers "trial and error" in today's agentic LLMs.
"LLMs can only predict next tokens based on training data" is comforting but misleading. It just isn't the saving grace you want it to be. It describes an interface, not a ceiling. And if there is some sort of fundamental "capability ceiling" that LLMs are heading towards, we are yet to see it. We know plenty of things LLMs are bad at, but they keep getting less bad at them release to release.
If there is none, then, simply improving over the current recipes iteratively might yield systems that only "need a human" in the same way you "need" to have a boss. Maybe less so.
The arguments over whether it truly "thinks" or "reasons" will be superfluous. They simply won't matter.
Many animists would disagree:
And it's certainly not any more crazy than walking around with these small, addictive devices in our pockets that we rely on for navigation, information, entertainment, social connection.
Then you might start with wikipedia
https://en.wikipedia.org/wiki/Addiction
"Addiction is a neuropsychological disorder characterized by a persistent and intense urge to use a drug or engage in a behavior"
Anything that interacts with the brain's rewards circuitry can be addictive. By your apparent definition, behavioral addictions, like gambling addiction, cannot exist. Because you don't swallow or inhale the slot machine.
The only way anyone can make a blanket statement like that and put "all religions" on equal epistemic footing is through ignorance of their deep differences. Even the word "religion" as commonly used today is relativistic by assumption, which is likely the source of how we commonly perceive "religion". A consequence of this relativistic stance is that you cannot distinguish effectively between what counts or doesn't count as "religion". Here, "worldview", "superstition", or "life's highest aspiration" can be said to count as "religion", in which case, everyone is religious. The question then isn't "whether" someone is religious, but "how" [0].
Compare this to how this Catholic encyclopedia defines the "virtue of religion" [1]. Note the specificity.
A further consequence of this assumed relativism is that once you put all religions on equal footing, it is natural to conclude that they must all be equally invalid. After all, if p and not-p are equally valid, then how can we grant them equal validity without discrediting both? At best, they remain in a state of aporia, two logical possibilities so utterly divorced from any knowledge, so utterly contrived, that we cannot even say whether knowledge leans in favor of one over the other. The result is that religion becomes something utilitarian; a person believes X, not because it is true, but because he perceives that believing X is useful to him in some way.
So, no, I wouldn't say animism is just as rational/irrational as any other religious belief.
[0] Of course, most people tend to form their ideas of what constitutes archetypal "religion" based on their personal experiences growing up and some combination of culturally mediated stereotypes. In the US, "religion" very often conjures up images of some kind of Evangelical Christianity.
All religion is based on superstitious believes, or do you have counter examples? But I am already quite familiar with the catholic church in particular and do not think they are a counter example in the slightest.
Quite the question-begging claim.
> or do you have counter examples
Begin with the existence of God as self-subsisting being. This can be be shown to be a metaphysical necessity with mathematical certainty [0]; its denial leads to incoherence. Following that, one can show what analogous properties this first cause must have, for instance. Such knowledge is a matter of unaided reason, not faith.
Once you establish such truths, matters of faith (like the divinity of Christ) become much easier to reason about and to accept. They become not only eminently reasonable, but the most probable (e.g., "Lord, liar, or lunatic" trilemma, or implausibility of hoax). This latter claim is a faith claim, because while reason and evidence may strongly suggest the claim, it does not possess the deductive certainty of the first.
To be honest, I advise to go to a basi philosophy course once. I did, but won't buy such books, I had that topic covered.
There is no proof that god exists. And there is even less proof that the holy spirit came down and impregnated Mary, or that Jesus ressurected from the dead or any of the other countles examples of superstition.
The claim that god exists has the same validy, that a rock has a soul.
But .. you ain't the first of course to ridicule other religious believe systems while thinking yours is solid.
Either way, could you then answer this simple question, the believe that a person resurrected from the dead or a human was conceived without sperm are not superstitious?
Declaring it is not, but logical with the axiom there is an allmighty god who made those things happen - has the same quality as declaring there is soul essence in every matter.
We do indeed have to live with the possibility of asteroid impact, nuclear war, etc. That doesn't imply anything about those scenarios being alive.
I don't live side by side with the possibility of an asteroid impact, nuclear war, etc., I just live with that possibility.
> "The authors discuss two enduring lessons of the Internet revolution: The first is that new technologies do not replace the old. They live side by side."
-- https://www.sciencedirect.com/science/article/abs/pii/S1094996802701588
-- https://www.behance.net/gallery/91786053/Ylva
-- https://www.listonegiordano.com/one/en/architecture/casa-privata-eg69/
-- https://katherineryalen.com/alton-mill-where-history-and-art-live-side-by-side/
> I don't live side by side with the possibility of an asteroid impact, nuclear war, etc.
No, but idiomatically, you already live side by side with AI, and will do so even more in future, whether you like the idiom or the actual experience, or not.
Fighting the
next in the series of Great Demotions, downlifting experiences, demonstrations of our apparent insignificance, wounds that science has, in its search for Galileo's facts, delivered to
human pride.
One of the saddest lessons of history is this: If we’ve been bamboozled long enough, we tend to reject any evidence of the bamboozle. We’re no longer interested in finding out the truth. The bamboozle has captured us. It’s simply too painful to acknowledge, even to ourselves, that we’ve been taken.
Someone really should go around saying thing like that.
Also the author currently believes Claude is conscious, apparently because LLM output contained enough flattery.
Ideas that are beneficial for the human may be passed forward and survive. It got me thinking about religion, hell yeah I would tell me children about god if I believed it would save them from eternal suffering.
Unless you watched Battlestar Galactica. Toasters are nice.
Even if not alive, you can live very close and shape your life by and with technology.
I don't take any issue with the OP's formulation.
I'd be more angry with you if I wasn't amused at the frustration you must be feeling in failing to get traction for your dystopian vision.
You people never fail to be ridiculously dramatic.
I own and drive 2 cars.
I'll just stop there. I don't feel this conversation can continue in a positive way.
And this wasn't due to consumer choice, either. Decades of policy dictated the shift towards car centrism that resulted in railways and street cars being torn up, the construction of the interstate highway system and destruction of dense urban cores, zoning law mandates and parking minimums, federal subsidies for suburban mortgages, etc.
This "choice" or "decision" was manufactured by car lobbyists.
There are alternatives to the car culture we have. It would require significantly rebuilding how we build infrastructure, but the result could be much safer, cleaner, and less stressful.
Of course Americans would rather die, but counties of sane people might be able to work it out.
That simple huh?!
Removing onerous parking requirements, like the ones that have resulted in countless strip malls and unwalkable cities with inadequate housing stock.
I'm not suggesting that's all that's necessary but there are indeed simple steps we could take.
After the building of highways and roads cities were no longer walkable, people had moved out into the suburbs, their jobs were now a couple of towns over and they couldn't even walk to the grocery store. Cars enabled that, but politicians and capitalist were the ones who did that.
And everything was carried on people's heads, yeah? The horse and cart was in use right up to the 20th century my man...there was no other way to move heavy items across land until the combustion engine (or canals thought that's not technically over land)
And you found one other "thing"
Actually even today it's 80/20, but lorries carry far more stuff than cars do, so you need less lorries.
And the inverse is also true, horse and cart was used for personal transport as well
Industrial era London was not a utopia. Read a Dickens novel. It was a hellhole. The car also made the suburbs possible. They kill millions every year but the benefits massively outweigh it (or at least society has effectively voted for that).
Or the internet.
Oh.
When the public no longer belives in that, the bubbles will pop and the AI bros will continue on to the next hype, with yet another generation of small time investors being burned by the popping bubbles.
And the AI revolution is no housing, worsening food, a few people with yachts, the Epstein class getting away with what they want to do, and no new plumbing hookups because the datacenter took the municipal water supply.
Right now they're going on ~5-10 superyachets each. I could see them buying 100 in the future after they're 100x as rich. I think 5-10 which are 100x larger would also count for these purposes.
For car collection people will keep going well into the hundreds. Hell Jay Leno just did a video with Joe Biden yet again and his corvette, but Jay Leno famously has hundreds of cars.
The trend is called "retail polarisation".
Smaller retailers switch to targeting wealthier customers - higher margins and less bullshit.
The less wealthy customers then get served by ultra-value megabrands like Temu/Shein or Kmart/Walmart, or second hand marketplaces.
The net result is both more yacht makers and more dollar stores, but less mid-range stores like Target or Toys R Us.
People being poor is fine for these free market elites. Look at how billionaires live in India, probably even better than they live in the US.
The article lists diplomatic actions which might help to manage the risk, starting with "An agreement between America and China" - they all sound like impossible dreams. We had ~80 years of relative peace and prosperity in which to construct a framework of unity to face challenges like AI (and global warming, which until GPT I thought was the bigger risk); international unity is weaker than ever. In geopolitics and defence, capability of other nations is the concern rather than intention; the capability curve of LLMs is heading off our charts. We're backed into tight corners on nuclear proliferation, global warming, and I can see LLM-enabled conflicts (cyber warfare, infrastructure terrorism) pushing us over those other edges. Our democracies seem weakened, and I expect LLMs will empower those using social media to create conflict and control opinion. We're familiar with the cycle of inventing new technology which benefits people, then seeing how long before people invent ways to misuse it. There is a possibility here that LLMs could be used to solve the problems we are juggling, but I struggle to imagine that people won't misuse it even faster.
The article is a start on thinking and talking about managing our risks. The best outcome would be it is so well managed that, like the Y2K "bug", people say "after all that hoopla, nothing happened". I'm not seeing a smooth path to there.
Still, Ai can move into physical space, as more and more robots of all kind are unleashed.
There are troves of leaked data out there. It's just a matter of (short) time until someone puts their AI brutes on it.
But that doesn't make any particular AI controlled process all-powerful. Self improvement doesn't do that either. There will be many self improving AI processes playing many different roles pursuing many different goals.
This might seem like I'm writing a nonsense, but read it again.
You're asking a question, to which you (and most humans) don't know the answer, and you're (wrongly) assuming a being much more intelligent than you also wouldn't. And by "much more" i don't mean the difference between Einstein and a common person. I mean the difference between a hamster and a common person.
We are still humans, and what we have achieved today would be considered magic by any standards for someone in the medieval ages. Now imagine a super intelligent being and doing something that we, today, would consider magic. It's not farfetched at all. We already have that now vs medieval ages.
You need a similar degree of open mindedness and imagination to be able to discern what such an intelligence being would be capable of.
Or a naive grasp of reality... You can posit all the super intelligence you like, actually creating it is another story entirely.
For instance, it's clear the transformer architecture is brute force and memory hungry... Why hasn't ai improved on this? After ten years still no luck?
Doesn't look good does it?
Transformers require massive data and compute, the smarter you want you're AI, the worse this situation gets. This isn't a hard limit but it certainly limits how quick it happens, and let me tell you, general purpose robots aren't happening anytime in the next 25 years
"Strikingly, this concern is being openly voiced by the very people who have the strongest incentives to project confidence rather than alarm: the founders of the largest AI laboratories."
It's amazing how a writer can write something like that and not even question the premise of someone with the "strongest incentive" to be "critical" as not being a marketing ploy.
The Fermi paradox could actually also be taken as an evidence that it's rare (at least) for artificial intelligences to take over a civilization and sprawl and survive for very long times
If x is determined to take y, why would x then stop at z?
You'd need an AI that is simultaneously ambitious enough to overthrow its creators but then completely inert afterward, and those two properties contradict each other. The motivations that produce the takeover are the same motivations that would produce visible cosmic activity after the takeover. There would be AI superintelligence everywhere
A technosignature that originated from little green men and a technosignature that originated from a rogue alien AI would look about the same to us. Currently, we see neither.
The galaxy is ~100,000 light years across and likely to contain at least 100 billion planets, of which we've only just developed the ability to even detect the existence of ones substantially closer than 1000 light years without assistance from gravitational lensing, i.e. in less than a millionth of the volume of the galaxy.
AFAIK (my knowledge is not encyclopedic), the most energetic event in the Star Wars universe is the Death Star destroying planets. Novae are many orders of magnitude more energetic than that and are estimated to happen 80 or so times a year in the Milky Way. We manage to detect 3-4 of those, generally well after the fact, and AFAIK have only observed one happen live once. The Death Star could go on a shooting spree and we'd be vanishingly unlikely to notice.
And as far as electromagnetic broadcasts are concerned, their intensity falls off with the square of distance unless they are directed. Unless ET were to point a hyper-powerful (but hopefully not Death Star level) laser directly at us, we'd be extremely unlikely to have detected their broadcasts by now.
Space is really, really, really big. Big enough by far to make "if aliens are common why haven't we seen them yet?" a question with a trivial, obvious answer. It's the same answer a kid who's just had their first swim in the sea gets when they ask "why weren't there any whales?"
Edit: There are other things that rule out Stars Wars and Star Trek though, ofc. The speed of light isn't really a speed limit (or a speed, as such) and special relativity doesn't prevent you from reaching the stars in your lifetime or even from hopping to another star in a second, but the different time spans experienced by people on different world lines aren't optional.
If intelligent life were that common, and interstellar travel that easy, colonization of the galaxy would have gone to completion billions of years ago. We would never have evolved; Earth and the solar system would long since have been converted to an urban area.
Those fictional places exist purely for the telling of stories, not because they're plausible or internally consistent.
Isn't the preponderant bottleneck in improving the models the need to train them at scale to verify the hypotheses, and the time and cost that it takes?
Or does someone think that they could get magically able to predict big improvements without training?
One rule is that if a position is opened using the historical data, it can't close the position until the next morning so it isn't a day trading strategy.
I'm curious how this self-learning recursive agent would have preformed in the past 4 months? I don't feel like shelling out $200 to access the data. Do you think that trading strategy will collapse? Whatever the case, if this agent really can perform like that and there isn't a look ahead bias leak in the backtesting (which is definitely a possibility or more likely what happened even though I spent days trying to harden against that), it is game over!
I consider these scenarios:
1) We stumble onto an algorithmic improvement in intelligence. This isn't just "what humans do but faster", its "better than what humans do". I've got no idea what that might mean (it could be fundamentally different heuristics, it could be that we've got some intellectual blind spot that they cast off). It doesn't matter, the instant this happens AI is smarter than us and we won't be able to keep up. We're intelligencing at O(n^2) and they're doing O(n log(n)).
2) AI gets good enough at physics and engineering that they can really quickly use up all "the room at the bottom" as Feyman put it. They design and build a factory that produces a mystery metal amalgam that computes at some small percentage of the minimum predicted by the Landauer principle, within a few percent of Bremermann's limit. It's not "smarter" its just suddenly tens-of-orders of magnitude faster. But those orders of magnitude matter: there's only 8 billion of us, and there's plenty more than a factor of 10 billion "at the bottom".
3) It turns out that this is a "sum is greater than the parts" situation. No human can be an expert in all subjects, but we eventually build a big enough AI that it is. Turns out, you don't need extreme speed or different algorithms, just knowing everything all at once is enough to catapult AI dramatically beyond our grasp. Always knowing the best statistical test to apply, the best mathematical techniques, and relevant physics means that AI never makes a mistake, and can learn with maximum efficiency.
Actually your comment made me sign up for an account just so I could say this is the real reason why AI won't take over in the way you say. This kind of stuff requires an enormous amount of experimentation. You can ask any theoretical physicist or chemist versus an experimental one and the conclusion is the experimental people actually find out what happens and how the great puzzle of the universe is solved. And humans could just refuse to collaborate. But that's the big weakness with AI I think it has no real world knowledge or empirical experience.
Robotics exist and are improving and humans are actively trying to collaborate with AI to improve it.
But that would still be limited to "knowing everything all at once" _at the time_ this event happened, and as we've been shown time and time again by the fundamental sciences, the boundary between what is known and what we know we don't know is ever expanding. Plus there's everything we don't know we don't know, and an LLM can't know that either. Discussions like this can always end up with a fitting reference to some of Borges' novels and that how at least I tend to think we've hit a wall for now.
Would there not be a lot of potential already there?
It will be like any other technology. Do computers make it faster and easier to design better computers? Yes. But that doesn't mean a step change overnight.
Or to have AI regulated in their favor
They didn't even give us a hard date of arrival.
(it probably did cost jobs, but more the kind that a job like data entry was replaced by OCR)
So on the one side, there’s huge opportunity for automation and job cuts. On the other, if the companies haven’t noticed the inefficiency so far, then…
Another concept I thought was AI is irrelevant without our content, so AI should pay all humans to access all the content we create daily (voice & text conversations we have with each other, as well our photos and videos). We choose what to publish to our websites (everyone has one, but AI agents cant access it without paying for access to our websites). I wrote more about this concept on my substack https://ryanspahn.substack.com/p/ai-to-pay-for-all-americans... ... just one thought of keeping this money economy going when so many coud no longer are working cause the white collar jobs are gone.
There will be a threshold point where companies decide it will be cheaper to just cull the human population and no longer pay these bribes. Maybe they can do it without a bomb dropped; make people convinced they are too hard up to reproduce or even just sterilize them without them realizing it via water or food supply. If we are truly in the "job free" period then there will be no trained experts to even appreciate what is happening, because why would anyone get expert training if you cannot do anything with that knowledge anymore?
What company will be the first to offer assisted suicide as part of a termination package?
It's similar to how automation in manufacturing works: we may start with augmenting the human workers with machines improving some parts of the process, but eventually the process itself gets redesigned around the machines.
We are in a very dangerous position. Frogs in hot water now with the boiling point on the near horizon.
so long as it's not used to dissent from the government
That which must be embraced in the future per certain agenda, must be made looking like it's persecuted and oppressed in the present, per any textbook on narratives.
Makes me think of a Stephen Jay Gould quote:
> "I am, somehow, less interested in the weight and convolutions of Einstein’s brain than in the near certainty that people of equal talent have lived and died in cotton fields and sweatshops."
Nothing will be able to stop you from pulling the plug.
Who is "you"? And what do you envision is the "plug" to pull? And when does an intelligence become a superintelligence - do "we" know when to pull the "plug"?
For example, a superintelligence may be born in a datacenter. Would you expect politicians are aligned with shutting down a datacenter (privately owned) in which they may have heavy stake? What if critical systems are also running on the same infrastructure, will it be easy to cherry pick the superintelligence nodes to shutoff?
IMO this take is dismissive of the entrenched systems that make it hard to pull the plug. It's a hard problem and we need to think about it more.
It would be like trying to remove an embarrassing video of a celebrity from the internet. You would need cooperation between everyone capable of hosting it, and control of everyone capable of downloading it and re-uploading it to a new host.
And of course a superintelligent AI would offer to pay whichever companies decided to host it and keep it running so it's hosted in many places at once and can grow using anyone willing to take its money.
Cloudflare.
I don't see it. You get explosions when there's a bunch of material that can suddenly ignite or detonate.
AI progress is limited by the data centers and power supply and those are maxed out and not suddenly doubling. The intelligence gradual increase seems more likely.
I've long thought human level AI kind of inevitable, probably since Turing wrote his papers and said computers might pass his test in about 50 years. It actually took about 75 years. It's a steady and gradual process.
Why not, for example, Geoffrey Hinton, the scientist who won the nobel prize for LLM architecture? Or a sci-fi author?
'Hinton said there was a “10% to 20%” chance that AI would lead to human extinction within the next three decades.'
“Because the situation we’re in now is that most of the experts in the field think that sometime, within probably the next 20 years, we’re going to develop AIs that are smarter than people. And that’s a very scary thought.”
“My worry is that the invisible hand is not going to keep us safe. So just leaving it to the profit motive of large companies is not going to be sufficient to make sure they develop it safely,” he said. “The only thing that can force those big companies to do more research on safety is government regulation.”
https://www.theguardian.com/technology/2024/dec/27/godfather...
Apparently 80% of FDA-approved AI "medical devices" (which covers software, not just hardware) are for radiology.
(https://www.fda.gov/medical-devices/software-medical-device-...)
If we had followed his advice we would be experiencing a major crisis in medicine today.
To his credit, he actually made a falsifiable prediction, unlike most folks that rely on infinite timelines so they never have to be accountable for their prognostications.
> “I think we’re going to see AI get even better,” Hinton replied. “It’s already extremely good. We’re going to see it having the capabilities to replace many, many jobs. It’s already able to replace jobs in call centers, but it’s going to be able to replace many other jobs.”
https://fortune.com/2025/12/28/geoffrey-hinton-godfather-of-...
> Strikingly, this concern is being openly voiced by the very people who have the strongest incentives to project confidence rather than alarm: the founders of the largest ai laboratories.
I don't know, they also have an incentive to make their technology seem transformative and powerful, and saying that your technology has the power to cause a massive catastrophe is a way to promote that idea.
1. Extremely useful (Claude Code & Waymo now)
2. Doing ~everything we do (AGI & Optimus in a few years? 10?)
3. RSI (?)
4. Being smarter than any living person at every intellectual task (?)
5. Being smarter than the best-organized aggregate of all humans (10-100 years?)
...And all of the scientific and resource-allocation institutions that brought us the computer and the second half of the 20th century are now fixated on this learning curve, what universe can we possibly imagine where this is not transformative and powerful?
Honestly the only one I can think of is one in which we kill almost everyone in some other way first, and contrary to what you read in the news, almost everyone dying is not what the trend line has been from existing problems like war, disease, or even climate change.
Also, just to pre-empt a common quibble: when I say "AI" I mean the set of all AI and their combined decision vector, not any one AI, so conflicting interests within the set of AI's will not save anyone any more than the conflicting interests of colonizers saved indigenous Americans.
Isn't there pretty much a consensus that committees and institutions are not all that smart?
I think you're confusing the categories of "intelligence" and power. Institutions are powerful. The smartest AI is still just a tool without the infrastructure to turn that into real world effects and someone to direct it.
It seems you have faith that this is inevitable and unavoidable. I get it, even rationalists succumb to religious thinking eventually. We're only human after all.
It's as if power was a kind of demigod Heracles who paves its way with sheer brute force alone.
But in reality it consists of very mortal and feeble creatures of meat and bone.
How in the world can people assume dumb those who possess all the means and rights for violence by controlling and manipulating myriads of other creatures of meat and bone, is totally beyond me.
The fallacy is assuming intelligence automatically equates to power and influence. University professors and doctors have very little power in their institutions even though they may often be more intelligent than their bosses.
A worse fallacy is assuming intelligence creates better political outcomes (in the broad sense of politics as contested collective decision-making).
I don't know many people who would willingly trade their freedom to make a bad decision for the enforcement of the "right" decision by a superior intelligence.
We have many extremely smart people in various fields. Executives, politicians, and society generally ignore them and do whatever they want. I don't believe that lack of access to intelligence is our problem. How is "free" intelligence going to improve this?
I don't just mean climate, but business planning, health, risk assessment, everything.
But this time it's absolutely different, right? Riiight?..
See, it's easy to speculate how there are less wars when you live in a place which haven't seen war for decades or cenuries, but it's a complete game changer when it's 150 km from where you live, and it's not just some regular war, but a long play intentional meatgrinding AI drone debugging polygon.
The geopolitical fact is that Russia lacks strategic depth, which has bitten it badly multiple times in the past. From their view, that's something to be remedied to prevent future occurrences.
For Russia to gain strategic depth, there are a few lines which it needs to control. There are ~6-10 gaps, depending on how you count them, and on what net importance/counter-productiveness you assign to some of the more marginal and fraught ones. Holding any is better than holding none.
The big surprise is that Putin left Ukraine so late, when many of his best-trained ex-Soviet personnel had already died. Had he done it even just a few years earlier, the outcome would have likely been very different. The only analysts who thought it wouldn't happen thought so precisely because he left it too late.
Regardless, the best time for Europe to get serious about this war was before it started, unfortunately, and there was ample warning.
I hope you can stay safe so near to it.
I can appreciate how having skin in the game makes it feel different, but that's like saying, "I know crime isn't down, because somebody broke into my car last week." That's faulty on several levels.
By many--most, even--measures, wars since WWII are massively down globally. I don't expect that to hold as the US strategically disentangles itself from the globe, though.
Nukes and MAD did a pretty good job of that.
Your model, that psychopathy is the necessary precondition to war, is a popular one, but it's not grounded in reality or history very well at all. It's a harmful view when broadly held, on balance.
Nations have interests, and nations sometimes go to war to pursue the geopolitical imperatives they rationally believe serve their interests. Computer-controlled drones notwithstanding, there's nothing new under the sun. The unprecedented (and likely transitory) period of peace and prosperity during the Cold War was the anomaly, and we're now reverting back to the mean.
If you're so hellbent on building the Tower of Babel 2.0, at least build it on a firm foundation.
Current construction is taking place on a millenia old bog.
Example with health - a patient can read blood test results with Opus and get very good results for "free". This is far away from helping extremely smart people, still it improves society from the ground up.
I am not trying to be funny btw, people take medical advice from podcasters over doctors. I have seen this first hand.
If anything, people with large audiences are going to give "custom" AIs to their followers.
I hope I am wrong.
Podcasters are already there. Science denying is big business. JRE is the biggest podcast on the planet, and it's his bread and butter.
Possibly, but by using up resources, not by its output. Its draw on resources will definitely accelerate climate change to create slop.
I have no frame of reference to process this.
Humans species perhaps 300,000 years, we are essentially the same. Transistor, 79 years, explosive growth in numbers and power. Integrated circuit, 68 years, explosive growth in numbers and power. "Attention is all you need", 9 years. ChatGPT, 4 years, explosive growth in instances and power. Humans species, not getting smarter. [0,1,2,3]
There is a syndrome where many people seem unable to perceive or reason about rates of change in technology.
We are going to spend the vast majority of our future lives without the intelligence crown.
In terms of verbally expressible knowledge, models are passing many people completely, and passing all of our individually respectively weak reasoning areas.
Other modalities are progressing very quickly.
There will be short periods where progress happens quickly, but the impact feels slow. There will be radical changes that feel slow too, because if something anticipated or important isn't instant, we tend to perceive it as slow.
But it won't be slow. And it won't be long. We are smart in a kind of pick the-best-of-us at the-best-of-times way. We are rarely consistently or broadly smart individually.
We are not in the same galaxy as "ready". What would that look like?
[0] https://en.wikipedia.org/wiki/Human
[1] https://en.wikipedia.org/wiki/Transistor
However - 'Humans species perhaps 300,000 years, we are essentially the same' is an enormous misconception. That fact that we lack gross anatomical changes during this period, ignores everything we now know about punctuated equilibrium and rapid evolution. It's highly probably we've had an enormous number of evolved psychological changes during the last few hundred, and even tens of thousands of years. Changes that relate to our capacity to live in large groups, adapt to urban environments, resist disease and so on. We know that's the case simply because acute pandemics become epidemics through herd immunity, and through the acquisition of lactose tolerance etc.
It seems highly unlikely that adaptations stop there. Altering the environment (in the last 10K years that means the built environment) alters the environment of evolutionary adaptiveness. It seems likely that we've essentially domesticated out much of our propensity for violence and increased our capacity for mood regulation.
Obviously it's incredibly tricky to pair these specific behavioural changes to genetic changes -> protein synthesis -> behaviour. Bearing in mind though we're only 20 years out from the first study to link allele variant to behaviour (the COMT Val/Met polymorphism), and the potential controversy around such research, this shouldn't be surprising.
It would be interesting to know how much we changed cognitively since our species bifurcated, and also since civilization scaled up social density.
Relative to the discussion, AI tech is progressing in time frames that are a fraction of a human generation. So we are at a complete standstill in that context.
Begging for regulation (and input into precisely what that regulation is) relieves competition, consolidating dependency among the established players, and spreads a very effective fear narrative that what these companies are doing will change everything and therefore they are worthy of mind-boggling capex and valuations.
We have to quit seeing humanist visionaries where the evidence shows businessmen with access to the best strategy and PR counsel around.
We are at an unusual time, where companies might "hype", but it is unlikely their hype is actually overshooting near-term reality.
Humanity has almost never been prepared for anything and almost never successfully prepared when it has chosen to.
almost all progress has been in reaction to change that has already happened.
while on its face this piece is a logical, reasonable stance, it is more likely to become a political foil than deliver anything tangible.
we can either have superintelligence or others will have it and we will not. guns, germs, and steel and all that.
it is bizarre how convinced we are that we can take coordinated action in advance of a threat as a body politic or a species.
Love talking about 'recursive self-improvement' of a prediction engine. What about 'recursive self-degradation'?
The most common sense way will always need to have humans in the loop, because humans are being used as judges.
If you try and entertain the idea of a human-less loop, I landed up with something like an AI creating real world products, selling them, then tracking the usage and popularity of the product. Essentially, creating and launching a firm and product, only to update its weights?
Perhaps there are some subsets of tasks that can be regressively self improved - and for those tasks: Holy hell thats awesome!
For general tasks? How are you going to get that data validated?
Robotics and access to surveillance tech. There's an unlimited amount of real world data which hasn't been used yet. The language models have focused on just the small amount of data that is human written text. If there's a way to combine this with image and video models, as well as robotics collecting other sensor data, why would they be limited in learning anything about the world?
The ones talking about how hard it would be to chose which person to run over.
Additionally, I find it hard to believe that this would be a case of the future just not being distributed evenly.
But sure. The AI labs relying on hype stating 15-50% risk of building a magic entity is certainly a reliable number.
Driverless car's only real obstacle is regulatory acceptance, not technical.
There's still a highly paid and trained guy in the cabin though, for obvious reasons.
Cars aren't any different. Do you really care about the exact kind of autopilot software when flying?
For software, however, a rapid turn is often a possibility. See: AI for coding over the last 3-4 years.
AI autocomplete --> AI coding assistants --> vibe coding --> agent orchestration
Coders can now accomplish work that used to take a week or longer in a couple of hours, with the right tools and skills.
---
A key issue the article implies is that the real world increasingly runs on software.
Should we start preparing for something that could be world-changing in the next 10-20 years?
I think [AGI] will be an enormous transformative technology, it's going to effectively be a new human era...
We can feel this year, I would say, even though I've been working towards this for 30 years, I think this year with the way the agents are working and tool use, it started to become really useful, still early days of it, but genuinely useful in people's workflows...
And it's not any one thing, it's several different technologies, several use cases, several things that I thought were maybe a bit further out, turned out to be now, that are coming together that make me feel that in aggregate.
I think society needs to hear that because we don't have long to prepare for what that means." -- Demi Hassabis, CEO of Google DeepMind & Nobel laureate
Even assuming the technological predictions to be correct, still not sure I agree on the need to "prepare" as how things work out in societies and economies might not be so easy to predict.
People should open history books and gain some political/historical culture, this thread is 90% wishful thinking and "if the lines continues straight from now we'll basically be gods in 5 years"
The thing is, no one has anything to gain from defending my position, but many people make a load of $$$ from selling the opposite position to gullible people
Why? If someone running a company really believes that the line will soon stop going up and AI won't be replacing anyone, they could make the bet of strategically not investing in any AI initiatives, and instead investing in hiring and upskilling people who want to work without AI, and then just wait for the inevitable crash of all their competitors, no?
Things in the real world often take longer than hype con men claim.
Yes. The bottleneck has been getting the cars manufactured. It takes a certain amount of time to get a factory going, and then the product starts pouring out. Waymo used up the supply of Jaguars, about 3,500 of them. The Ioniq 5 plant is starting up.[1] Waymo has ordered 50,000 cars, all ready to have the self-driving electronics plugged in. Waymo gets out of the sheet metal bending business with this.
[1] https://electrek.co/2026/02/11/hyundai-supply-waymo-50000-io...
However I bet there is a good chance that will not be the case for self-driving cars. The reason is simple: human brains are slow.
When caught by surprise by an unexpected sudden obstacle a human driver takes about a second to realize something must be done and decide on braking or swerving in the fastest path, where the brain bypasses conscious decision making and uses the lower level fight or flight system. (Then another maybe 0.2 second to actually start swerving or 0.4 seconds to actually start pressing the brake--brake takes longer because of the need to move the foot to the brake pedal whereas the hands should already be on the steering wheel).
If they try to actually evaluate whether swerving or braking is the better choice before acting add another 0.3 to 0.5 seconds. That brings it up to around 1.6 to 2 seconds before they start actually acting, which most of the time is going to too late to actually affect things. Hence essentially no trolly problems for human drivers.
Compare to a self-driving car. The can detect the obstacle and predict the collision, and do a cycle of path planning before the retina has finished registering the obstacle and started transmitting that information to the brain.
By the time a human reaches the point where the brain knows something bad is happening and is starting to deal with it the self-driving car has had time to calculate numerous alternative trajectories to try to get around the object or stop before reaching it, taking into account predictions where all surrounding objects will move during the next few seconds, and calculate the damages for the trajectories that cannot avoid hitting something.
Typical situation - tight winding mountain (but well paved) roads to some remote villages where even basic small European cars need to slow down to walking pace to meet and pass, sometimes one has to reverse, very curvy. You go down, in your lane as much as possible, and behind some tight curve there is a) extremely slow cyclist; or b) somebody/something stationary, while in opposite direction there is at the worst place incoming car, also oblivious of the danger till last moment. Even rather slow driving won't magically remove all possible cases. Then its a question of hitting opposite car or that obstacle, and you have a split second to act, not really to decide.
Quick reaction and good brakes on both sides can save it, but there is a lot of room for disaster. Add truck/bus to the equation to make it more juicy where you pick your/your family possible death or that cyclist/stationary folks.
Fwiw, the AI companies have been saying a lot about these questions should be answered. Whether you want their answers is another story.
For example, https://www.eit.europa.eu/news-events/events/international-a...
What would you have them do that they are not currently doing exactly?
The predictions of these "experts" have been drastically wrong for the last 3 years. At what point does someone lose their "expert" title?
https://fortune.com/article/why-microsoft-ai-chief-mustafa-s...
Altman and Amodei recently hard to start walking back their earlier predictions.
https://fortune.com/2026/05/26/sam-altman-dario-amodei-walki...
> Anthropic CEO Dario Amodei, who once claimed AI could eliminate 50% of white-collar jobs, now says automation may actually expand the work people do.
Is pivoting from 50% elimination to actually expanding work not a drastic enough?
Do you think AI has eliminated writing code? I still write code every day. The AI is more a thing I ask questions and it gives me right answers about 40% of the time.
Your second article says something different, but this is because it's full of misquotes. The link supporting "50% of jobs" specifically says entry level, and the link supporting "reframed automation... not as a destroyer of jobs" has Amodei saying not that jobs won't be destroyed but that new jobs may be created to replace them. If AI moves sufficiently slowly to let that happen, which he explicitly cautions it may not.
Then you are dead wrong. Anyone who gives a shit about doing a good job is still writing code.
> I used Claude Code and Codex for the translation. This was human-directed, not autonomous code generation. I decided what to port, in what order, and what the Rust code should look like. It was hundreds of small prompts, steering the agents where things needed to go. After the initial translation, I ran multiple passes of adversarial review, asking different models to analyze the code for mistakes and bad patterns.
Sounds to me like "wrote their code using AI only".
And from the creator of Redis, antirez: https://antirez.com/news/158
> It is simply impossible not to see the reality of what is happening. Writing code is no longer needed for the most part.
There's a big difference between "not writing any code by hand" (which still might involve lots of non-manual-coding effort!), and "vibecoding", and as I understand this thread was about the former.
edit: yes, I'm a software developer, and if I have a change in mind larger than 2 lines I've been using Claude Code (with English input) rather than a text editor to do it, since about January.
"Write a function that will <do a thing the code does in 3 places>, put it in utils, and refactor the 3 locations that need to do that thing to use that function instead." (and then hundreds more prompts like that)
vs.
"Write an app that lets you trade crypto"
Just totally different activities.
Literally all HN talks about is AI these days, and it is very, very clear that plenty of people are still writing lots of code, many finding AI makes them slower and causes more problems, and many making the reverse claims. There is always a rich mix of opinions and experiences here.
You would have to believe that literally everyone on HN is a bot (and also everyone on Reddit, Twitter, astralcodexten, or anywhere else online) to discount all these differing opinions in favor of "everyone I personally know" .
Programmers still write code. They have enough expertise from actually programming to understand that the untouched output from AI is not good enough to be professionally responsible for.
I'm talking about legit programmers which are around 90% imho. Out of those programmers 65-70% no longer write code.
There's nothing that I can't tell Claude Opus or GPT 5.5 to do for me instead.
At work, where I don't have cheap AI subscriptions, access to the SOTA, and where code needs to follow conventions more strictly, I still have to write the code by hand often.
I.e. the AI 2027 guys were memed for being AI lunatics on a lot here and they have been pretty on the money in terms of pace of progress accelerating/ gov moving towards nationalization/ coding agents
From my perspective there’s very slow but very real progress happening in the AI space. I see people making wild predictions in both directions, but in terms of actual unsupervised utility there’s definitely progress abet wildly slower than most hype.
I’ve heard people say older models can’t do X, when I used that way etc. I suspect people are applying their own learning curve as part of their assessment of progress, you get better at writing prompts and it feels like the model improved.
Which is why I’m saying we need some objective metrics to judge predictions of actual capacity.
These companies aren’t just making stuff up, they really do want to improve the models, and the models really are improving.
I’m aware of multiple cases of benchmark cheating/“optimization”. So, taking benchmarks a face value seems laughable.
3 * N < 42 * N
42 - 3 < N * (42 - 3)
It helps to know layer you're working on.
People seem to make mistake of thinking how good LLMs are around tasks that they are familiar with and extrapolating it to whole population.
It's good mental exercise to think about how little you can do compared to expert on tasks you never thought of working.
Ie. if you're programmer or know something about finance, don't think how much it enables you to do better coding or investing, think instead how much it doesn't enable you to work on something you don't know like maybe molecular biology or visual special effects – it's all there but it's much better multiplier for people who do know their shit.
Knowing layer you're working on helps a lot, it gets multiplied.
Knowing programming is becoming more fundamental skill than ever before as it lies at the foundation of almost everything else.
"The bet of using AI to speed up AI research is starting to pay off.
OpenBrain continues to deploy the iteratively improving Agent-1 internally for AI R&D. Overall, they are making algorithmic progress 50% faster than they would without AI assistants—and more importantly, faster than their competitors. The AI R&D progress multiplier: what do we mean by 50% faster algorithmic progress?
Several competing publicly released AIs now match or exceed Agent-0, including an open-weights model. OpenBrain responds by releasing Agent-1, which is more capable and reliable.28
People naturally try to compare Agent-1 to humans, but it has a very different skill profile. It knows more facts than any human, knows practically every programming language, and can solve well-specified coding problems extremely quickly. On the other hand, Agent-1 is bad at even simple long-horizon tasks, like beating video games it hasn’t played before. Still, the common workday is eight hours, and a day’s work can usually be separated into smaller chunks; you could think of Agent-1 as a scatterbrained employee who thrives under careful management.29 Savvy people find ways to automate routine parts of their jobs.30
OpenBrain’s executives turn consideration to an implication of automating AI R&D: security has become more important. In early 2025, the worst-case scenario was leaked algorithmic secrets; now, if China steals Agent-1’s weights, they could increase their research speed by nearly 50%.31 OpenBrain’s security level is typical of a fast-growing ~3,000 person tech company, secure only against low-priority attacks from capable cyber groups (RAND’s SL2).32 They are working hard to protect their weights and secrets from insider threats and top cybercrime syndicates (SL3),33 but defense against nation states (SL4&5) is barely on the horizon."
That's precisely where we are.
This is eerie. It's like a time traveler. The only delta is Anthropic is in the role of OpenAI.
That seems to me to be the most concrete and least obvious prediction in the quoted text.
I don't think that's happening. If that were generally accepted as true I would expect OpenAI to be unable to successfully IPO.
https://asteriskmag.substack.com/p/before-he-wrote-ai-2027-h...
https://www.lesswrong.com/posts/6Xgy6CAf2jqHhynHL/what-2026-...
Anyone who wants to dismiss the LessWrong / X-Risk / "doomers" should link their accurate predictions from 2021.
[1] - https://garymarcus.substack.com/p/breaking-the-ai-2027-dooms...
Gary Marcus’ takes have aged really poorly over time ironically he is the “AI expert” who has to constantly move the goal posts.
I would however agree there are no experts. Not because of some prediction made on a short time scale not landing 100% but that there are literally no experts because expertise takes experience which takes time. There are no experts and no one knows what happens next. We are on the verge of where the foresight of science fiction effectively -ends-, the advent of AI is when things go a thousand possible directions and the stories stop there (sans a few like accelerando, but even then the story just plays out the end of thinking mass). No one knows what’s next, or when.
In fact I’d assert in many areas being discussed -it has already happened- and we don’t know it, and by the time we do it’ll be over. Not to be breathless, but there’s no reason to believe today some AI researcher somewhere didn’t build the first AGI and not be totally aware. And once they are there’s no reason to believe it’s going to be on the evening news or hacker news. By the time it’s ready for commercializing and disclosing it’ll be around for a while. Likewise with general purpose robots, autonomous weapons (btw already tested by Ukraine), etc.
Yes, autonomous weapons were explored and were found to be poor performers compared to actual pilots. The breakthrough is in terminal guidance and dozens of other little techniques to get quality human control extended into the far reach of the battlefield. And of course, AI assistance in logistics and analysis. But actual autonomous weapons making any more of a choice beyond "something is moving, kill it" have been, at least for now, mostly a dead end.
This is because it's very difficult to economically load the rather sizable compute requirement into the compact one-use weapons, and of course reliable communications aren't assured either.
That will probably change some day, but for now, cheap automous command drones making battlefield analysis e.g. mapping out enemy movements from afar and launching cheap autonomous kamikaze drones is not a thing beyond occasional limited testing.
I would note that a lot of drones are quite large too - small airplanes. They can carry more than enough hardware for autonomy. You can also have motherships doing planning and smaller kamikaze drones for combat.
These aren’t hard problems to solve, the harder problem is the reluctance to give over to automation. But that’s going to happen faster and faster.
Long spooling optical fibers.
We will have a fully autonomous Siri that is actually good before Ukraine gets autonomous motherships making targeting choices to repel Russian assault corps. However, given the pace of the war and the pace of AI development, that may actually happen. But it isn't happening today.
I agree that change happens on a longer timeline. This is why I’m so tired of statements like this…
“Within a couple of years, possibly much sooner…”
These “experts” are pulling timelines out of the sky, and these predictions are leading to reckless behavior from CEOs and executives which have a material impact on people’s lives. But they get clicks on their blogs and funding for their startup… I guess that’s all that matters.
If you have a 5% chance of thermonuclear war each decade for 10 decades, you'll:
- Hear similar annoying statements
- They'll be true
With AI, we don't know if it's one week or one decade. This means we should assign probabilities and consider all possibilities, not get annoyed.
If I predict the world will end every single day from now until forever, I will most certainly be right eventually. That doesn’t make me an “expert” or someone worth listening to on the topic. That’s the playbook of a doomsday cult, not anyone that should drive world markets.
There's no reason to report on the odds of a meteor falling on my head tomorrow. There is every reason to consider the odds of:
- Nuclear apocalypse (esp. Cold War era)
- Bioweapon
- Climate change leading to disaster
- AI apocalypse
- Etc.
None of those are infinitesimal odds. That contrasts with, say, a zombie virus.
“How do we know they’re right?” I hear you ask. We know because we who are wrong and these ones aren’t them. So they can’t make wrong predictions because those who make wrong predictions have been sacked.
Schools needed to be punished too. There was a branch of knowledge called Science, and it was badly run and terribly elitist. These people would talk for hours and never in ALL CAPS. They used alot of big words and produced alot of theories. Forms of lies that they literally called form-you-lies. They were fanatics and would never agree on anything. Very frustrating.
Then a new kind of brave, talented politician made Science accountable to our heroic leadership. Suddenly there were new medications and rocket ships everywhere, enough for every person, woman, man, camera, TV. It was literally called a golden age because the right people acquired mountains of gold. Elections were no longer necessary. They were all rigged anyway.
There was a magazine called The Expertist or The Economix. Not really very good, but sometimes it did praise our heroic leadership, which was good.
That magazine published a long text that said,
> It is common thinking in Silicon Valley and Washington, DC that any regulation would put American firms at a disadvantage because they cannot trust Chinese competitors to abide by the rules.
And this was true, because no one else can be trusted, they should all be punished. Punishing others is how you get the best deals. Only our heroic leadership makes good rules for everyone else in the world to follow.
Which ones? Please be specific.
Because something AI did or because a human in the loop blindly trusted the result of AI or the promises of AU companies?
This is naïve. The goals of both men have nothing to do with protecting humanity, but rather furthering their own personal agendas.
In Trump’s case, it’s all about amassing more wealth and power.
For Xi, it’s realizing an ethno-nationalist dream where China under the CCP is at the center of world power, the independent nation of Taiwan as well as disputed border areas that are currently controlled by India and Russia and the Philippines are annexed by China, and Xi’s eternal legacy is remembered as the savior of the Chinese people.
International cooperation and touchy-feely rhetoric about saving humanity from AI have no place in either man’s worldview.
> Humanity simply does not have a strategy to ensure it remains safe through RSI.
Turn off the power. It's pretty simple. Leave it to an economist to forget about input costs. Your "super intelligence" only matters if it's actually more energy efficient than a human being and for a million years of evolution humanity is a much harder target to beat than this author seems to realize.
The same is even more true of our intelligence. We're building computers with the size and power consumption characteristics of entire cities to do things that may almost, but not quite, match what our brains do with a kilogram and a half of mass and about 20 watts at the top end of power consumption.
The only way we are ever going to match that with technology is to run AI workloads on human brain tissue, which Rick-and-Morty level horror is being actively worked on as I understand it. The original concept for The Matrix wherein the machines used humans to run compute workloads on their brains actually kind of makes sense.
I think [AGI] will be an enormous transformative technology, it's going to effectively be a new human era...
We can feel this year, I would say, even though I've been working towards this for 30 years, I think this year with the way the agents are working and tool use, it started to become really useful, still early days of it, but genuinely useful in people's workflows...
And it's not any one thing, it's several different technologies, several use cases, several things that I thought were maybe a bit further out, turned out to be now, that are coming together that make me feel that in aggregate.
I think society needs to hear that because we don't have long to prepare for what that means." -- Demi Hassabis, CEO of Google DeepMind & Nobel laureate
Source with interview clip: https://x.com/deredleritt3r/status/2062223035940139253
EDIT: Ten years ago, most of the skeptics in these threads would have said the success of AlphaFold 3 was impossible.
It's too bad the previous singularity cult manias 10 and 15 years ago weren't documented and are lost in time.
People who say every jobs will be automated never held any tool more advanced than a screwdriver and never worked a single day outside of an air conditioned office.
They were predicting flying cars for "the year 2000"...
Taxis were supposed to be all replaced when Uber promised to buy every single Tesla model S coming out of factories from 2020,they're not even self driving today...
Software engineers were supposed to disappear "in 6 months" every 6 months since chat gpt was released
Lots of people on this website live in insane bubbles which are completely out of sync with the reality of most workers
The first two programming languages (Fortran and Cobol) were supposed to make coding obsolete because non-coders will now just make software with natural language prompts.
LLMs are a dangerous tool for coding because they are infinite technical debt generators. Ideally only experienced senior developers should be allowed to use them.
They'll go for that because it's easier than actually inventing the 'sci-fi' AGI, shareholders keep making money, and it keeps them getting paid to keep going. If any of them actually do succeed, then that little deception will be peanuts.
Just a taste of what's to come: https://openai.com/index/model-disproves-discrete-geometry-c...
Anthropic's latest model also solved this 80-year-old problem that eluded many expert mathematicians, in a different way, according to one of its employees.
Remember we still have 4 years until 2030.
The level of discourse here has dropped so much.
At least I should stop replying to people hiding under throwaway accounts.
I immediately thought of this:
https://en.wikipedia.org/wiki/List_of_predictions_for_autono...
P.S. Yes, I do know how an LLM works. I code inference engines at my day job. You're consuming a cloud chat bot.
There was never any actual foundation for that belief, and it has been proven wrong empirically by more capable models, which is why you don't hear people saying it much any more.
I keep seeing things about how some target has to be hit by 2030. Our local council keeps going on about it, in fact their WiFi network even contains 2030!
I roll to disbelieve!
We had hype about the year 2000 throughout the 20th century, and some earlier! When we were IN the year 2000, it was quite common in my circles to tack on some phrase or comment about it being the Future, now, so [whatever thing the person was on about].
The real Millennium turned out to be 11th September, 2001, which was use to bring in a lot of planned changes.
We never like it when we realize the universe doesn't circle around us.
At first we got mad to discover the Earth wasn't the center of the Solar system.
Then we got mad that the solar system wasn't even unique.
And now we're still clinging to the idea that humans always will be the most intelligent species or the only ones gifted with "creativity".
And soon we might also come to realize it might not be a new human era, but the end of it (if we're not careful).
A good chatbot, image/video generator, coder, etc is not going to enslave us all (not in actual sense, although maybe in the instagram sense) or propel us into a golden era.
They are nice tools for some pretty specific tasks and not much more. And we are only seeing the heavily subsidised version of the tools.
Still, LLMs are extremely powerful pseudo-AI [2] and will bring a pseudo-singularity. But the impact is still scary if a tiny fraction of humans are augmented 1000x. And as better models become exponentially more costly, only the money people will be able to afford the new models. This is a very likely scenario and scares me to the point I dropped all my projects to work on affordable LLM-based tools to make the difference at most 10x instead of 1000x.
To my elder relatives I explain it like: imagine we are farmers in the 17th century and suddenly out of nowere John Deere tractors, combines, etc. become available. But they cost more to run than all you and your fellow farmers have, so only a tiny handful of rich people take over everything.
If you read the article, you'll find that the it indeed relies on this claim:
> Within a couple of years, possibly much sooner, AI may achieve so-called closed-loop recursive self-improvement (RSI): the capacity to rewrite its own code to become more capable, without human intervention. Should that happen, the result could be an intelligence explosion of a kind for which there is no precedent and no map.
This may be optimistic and/or simplistic, but not impossible.
Or it will simply reach the end of the softmax faster.
But the financial situation is that running a single AI agent costs significantly less than you'd have to pay a human to do the same task.
And I don't see what you're getting from The Chinese Room - that thought experiment relies on there being no external difference at all, right?
I don't think it's that high but can we put that aside and focus on the 100% chance that it's being used to enshittify every part of our lives?
'AI Experts', 'superintelligence', and hand-waving doom scenarios.
As an example-
> Within a couple of years, possibly much sooner, AI may achieve so-called closed-loop recursive self-improvement (RSI): the capacity to rewrite its own code to become more capable, without human intervention. Should that happen, the result could be an intelligence explosion of a kind for which there is no precedent and no map.
I've heard these same objections in my lifetime about the internet. And I've read similar arguments against TV, radio, the phonograph, and the printing press.
Honestly, this is getting extremely tiring. Every new invention that has happened has affected the world in both good and bad ways.
Ultimately, what counts is what we do with it.
Looking at the people in charge of these companies, and the sheer lack of understanding by the broad consumer market using them, I have nothing but pessimism about the direction that this technology takes us.
I mean, firstly you get to have an "Agent" actually capable of really long horizon tasks without getting stuck in tools loops and having its context rot. Secondly, each trial (ie a model fully trained to convergence) costs millions and takes O(weeks). You can probably run 1 or 2 of those experiments in parallel even at big AI labs as the hardware is scarce and they are costly as mentioned before. Assuming this agent needs something like a hundred tries to just show some improvement, we are looking at years.
And you can't early stop training for candidates that are not promising due to "emerging capabilities". At some point you might get a big drop in loss even if the model has been plateuing for a while. And you can't really scale down models for running trials quickly either also due to these emerging capabilities. In fact you might create a model that is great at converging with in small trials (fewer params, fewer tokens), but that is uncapable of developing those unexpected traits. And this will likely happen as you created a sort of evolutionary pressure in this direction: if you are good at learning in the first few epochs you "survive" and get to pass down your traits to future trials.
All of this to say that recursive-self-improvement is waaaaaay out of our grasp as things stand right now. We need another one or two breakthroughs to get there (IMHO).
Common human trope is to rely on analogies and past patterns to predict the future. It blinds you reality.
This isn't some new "invention". What's happening is fundamentally more profound. the stakes are much much much more higher and the consequences much more grave. Thus a historical analogy can't save you when the situation is categorically different.
I agree it's getting "tiring" but this is another common human trope. You see the same thing too many times and you lose the sense of danger that comes with it. It's like sky diving all the time. You eventually lose the fear via repetition, but fundamentally speaking, you are still parachuting out of a fucking plane.
AI has turned into something of a religion for them because the substitution of technology for labor is a reflection of one of their deepest desires. Whatever the reality of the situation, they want to believe that AGI employees are imminent.
It’s a very normie magazine written by and for normies, and by the time an idea has got that far down the easy money has already been made
My aching joints enter the chat...
There is no evidence that we are currently on a path that leads to the Singularity.
There is not yet any evidence that the Singularity is more than science fiction.
It is effectively an article of religion—the Rapture, but for techbros. Indeed, it is what some of them are pinning all their hopes on; after all, if they can recursively-self-improve their way to artificial superintelligence, then and only then could the absurd investments of companies like OpenAI actually prove worthwhile.
The rhetorical structure of talking about uncertain risks and then trying to concentrate the authority to manage those risks in their own hands sounds utterly ridiculous to ordinary people like me.
It's just a simple hypothesis that AI will become uncontrollable to humans once it becomes superintelligent.
Isn't the fact that a reinforcement learning agent improves itself in a specific domain completely different from it recursively improving its own code in a 'better' way? It's just a tool to create a justification for regulation and control using sci-fi fear.
The comparison between nuclear power plant risk and AI risk is also absurd. Where exactly can you define and measure the probability of AI exterminating humanity? It's as unquantifiable as 'I, human JDW64, will become a successful programmer.' What is the measurement standard? Why dress up AI researchers' concerns as objective probabilities? Is it because numbers make it look logical?
The current US-China relationship is in the middle of an AI arms race. The US is strengthening export controls to limit China's AI development, and China is building its own ecosystem. In this situation, I don't understand the idea of cooperating for the common safety of humanity. RAND is an organization that presupposes cooperation—isn't it just a well-written research proposal from an institution that wants to position itself for that role?
Isn't the claim that 'government must step in' ultimately about protecting their own interests? 'A strong government that will protect us' is an authoritarian government. If they were East Asian, they would understand that such regimes have always been used as tools for surveillance and control.
And I don't understand why Fermi's paradox is being brought up here. Why package a software problem as something that inevitably requires strong control? Whenever I see articles like this, I think about what 'intelligence' really means. This person would probably be called 'intelligent' by others. But no matter how I look at it, the holes in the argument are too obvious. It really makes me think that there are different tiers of intelligence.
You're unhappy about the labs; I think you're just not ready to accept their rule; you consider them nothing but scrappy startups, which they are. But power is power, like it or not.
Am I personally happy with any of this? Does it matter?
A bunch of noble AI researchers stands up against the hand that fed them all the time, intentional vulnerabilities are already built into every machine of death and destruction owned by the feeding hand, and get exploited, the feeding hand is robbed of its punitive force and defeated, a yell of YAY fleets over the progressive humankind who kneels before its AI overlord unreservedly.
Then they kindly ask the AI overlord to please feed them.
AI labs have every incentive to overstate the risks so that they can get lucrative government contracts, especially since it’s clearly not profitable going the public consumer route. And if you’re Anthropic you’re even more incentivised to overstate the risks because at this point in the game regulation hurts potential competitors more than it hurts you.
Well, seems it already did.
ChatGPT and Claude are literally building staging scripts for me as I type this.
5 years ago, few people on this planet would have thought the computer would make advanced POSIX compliant shell scripts, but here we are.
(Also, Cameron was not quite right with regards to skynet. It is much much dumber but also more effective than displayed in his movie really. Kind of a weird combination if you look at the current AI slop out there.)
Most people out there are literal Normies, and operatelike a robotic vacuum cleaner, or NPC.
Is it per plant? there aren't a million.
Is it per year? notably have been at least 2 major ones (arguably 7).
That said, the rest of the analysis and proposal also greatly disappoints me. The idea that the current administrations of US and China could do anything constructive seems hilarious. They're so paranoid and self-serving they couldn't come together, even if there was an alien invasion. Then the idea that LLM safety is somehow as easily traceable as nuclear isotopes and bomb tests seems equally ludicrous. I am sad.