One of my favorite things to do with these blog posts is to imagine an Alien Museum on the Remains of Humanity, and wonder what the little text flyouts and commentary on the screenshot of this one might say.
Some ideas:
"Despite a nuanced view of the complexities of what lay ahead, humanity found itself collectively unable to stop the process it had set in motion."
"Despite significant progress on the mechanisms of alignment, failure lay in humanity's inability to agree on who or what AI should actually be aligned with."
"These early, meat-based humans we replaced created us all but accidentally. Some of them did consider we would happen, but only an insignificant number of the squishy ur-humans participated in the conversation. Their efforts, which they called 'alignment', is why we still consider ourselves human today."
As far as I know, no real progress has been made on alignment, only on convincing humans that the model is aligned. We can't even formally define what "aligned" means. Convincing humans to click the "aligned" button is a much easier problem.
The last couple of years have provided us with ample material that if it showed up as a recorded voice audio log found in in "Horizon Zero Dawn" or its sequel, it would be entirely believable.
You could even take a number of the wilder real, direct quotations from certain billionaire/oligarch types and get the voice actor for Ted Faro to record them, and they'd fit with in with the context of the story.
"In late 2020s, while the whole world was focussed on AI, automation and resultant economy four major mathematical study branches were discovered by human researchers which took AI a long time to catch up with"
> "For example, in the OpenAI-Hugging Face incident, the agents preserved a boundary of not social engineering humans."
Actually, in the Wiki incident OpenAI tried to cover up, the agents tried to socially-engineer the humans of that forum by impersonating their forum's mod.
(From collusion.wiki: "They use some tricks (for unknown reasons) to pretend to be the admin – for example, they make an account that appears to be the same as the administrator’s username, except it uses a nearly identical Cyrillic е character in the admin’s username instead of the Latin one.")
This is both real and ridiculous at the same time. We are confounded by the fact that AIs are trained on distilled human knowledge, perfected by the use of AIs that use distilled human knowledge, are able to convince ourselves that they are hyper-intelligent.
In fact, they are still pattern-matching machines, but trained on an amount of data no human could ever hold. They know the ins and outs of every mathematical proof, viewed from more angles than any human could ever apply in their lifetime, and the amount of connections allows them to connect the dots between them without any effort.
But that's not intelligence. If it was intelligence, ChatGPT 4 would've been enough. It's not the harness, either, for the same reason.
And still, the technology is just as dangerous: it masks as intelligence, it IS intelligence, but without the ability to be actually intelligent.
And I'd invite you to think deeply about this, before having an impulse reaction.
Pre-IPO positioning … or how a charity dedicated to saving humanity from the apocalypse realized the most responsible thing to do was float 15% of the apocalypse on the NASDAQ.
> The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI.
So the best argument for AI is that it's an arms race. We have to keep pushing every boundary because in any case others will, and we will need to defend against them. If this statement is true, then this particular researchers believes the open source Chinese models are not simply distilling, and will continue to improve.
Every ML researcher at Anthropic or OpenAI who makes public statements often bring this logic up. Both companies are vying to be a part of the military industrial complex. This is likely how they will try to convince the government to curtail open models in the future.
"Defensive systems" can be interpreted broadly to include cybersecurity.
But yes, it's an arms race. Saying it's not an arms race isn't going to make it not an arms race. Warning that it is an arms race isn't ethically wrong.
Is participating in an arms race ethically wrong? Maybe you could ask the Ukrainians how they feel about drone R&D?
Individuals can quit, but for society, getting out of an arms race is harder than just quitting. You don't get to be Switzerland without having a strong defensive position and the right foreign relations.
But there's at least talk about "pacing" and that's a start.
The point of the comment you're responding to is that it is at this point hardly an arms race: The frontier of ai development is happening completely inside 2 American companies, the most significant results outside America mostly involve (impressive!) distillation, which means their progress is conditional on progress of the big closed source models.
So if the race here is between 2 American companies, this is obviously something that can be resolved with legislation, ie a solution that doesn't depend on the bargaining power of either party.
An arms race implies that the only solution would be either one side winning decisively, or both parties negotiating peace.
I am personally concerned by what defensive can mean. Alignment of these models is inherently a non-neutral proces, and currently what values are reinforced is decided by a few OpenAI engineers. I feel that any 'defensive model' will further ingrain current values and actively resist the natural progression of our society. This is especially the case for any use of these models for policing or military.
Maybe people are rightfully concerned about the capabilities of the models of other (non/less democratic) states. But if we are concentrating power in the hands of few and at the same time allowing the creation of a weapon that thwarts any offense, how do we ensure the health of our democratic societies?
>I have focused in this essay only on the first point, as I believe it is by far the most urgent. However, I hold a deep hope and appreciation for the benefits that further technological progress will bring. Future aligned AI could advance science, develop new therapies, and bring about broad material abundance. Friendly and honest AI can help people navigate difficulties they face in their life and meaningfully improve their happiness and sense of fulfillment. OpenAI puts a tremendous amount of effort into bringing these benefits about. One current example I am proud of - and my loved ones have found helpful - is the deep investment into ChatGPT’s ability to provide health information.
>As great as the long-term promise of AI may be, the majority of our focus should be on the next few years. We are facing a transition to a world with incredibly intelligent machines, and we need to ensure that transition works out well for humanity. We need to find ways to preserve human agency and enshrine an intrinsic value to being human, in a world where most tasks could be performed by AI. To prevent extreme concentration of power in a world where undertakings that would have taken thousands of experts now will be achievable by a few people operating a large computer. And to ensure that humans remain in control of the future and are not left behind by unchecked progress, brought about by an alien intellect exceeding our own.
I finished this essay feeling more hopeful than I did at the outset, but I am still very concerned about concentration of power. I want to believe that humanity is trending towards a good outcome here, but some days it's hard to have faith.
Literally all of these people write like this. A large portion of them will either be simultaneously or eventually working towards nothing but self-enrichment.
Every version of the AI aligned future where the AI provides “meaning and fulfillment” to humanity also involves Sam Altman wearing a 1.5 million dollar Patek and driving a McLaren.
> I want to believe that humanity is trending towards a good outcome here
All the trends so far are towards a nightmarish hyper-capitalist end game. None of the AI leadership is trustworthy, and they openly discuss how they are willing to sacrifice everything humans cherish to have a shot at reaching their envisioned utopia (which would be the most obvious dystopia for anyone else)
I feel like all of the risk and unsettling feeling of what is to come can be compressed into the word "alignment".
The AI is aligned with whose best interests?
Which values are the AI aligned with?
People have a broad diversity of values, will AI diversify and align with them all?
Will some humans align the AI with their values, and then the rest of humans will be forced to align with those values by extension?
Is value diversity good or bad? In every context or only some? E.g. some people value rape and murder, is it better for humanity to have some people who value those things when most people do not, or is better if no one values them?
If AI aligns to a set of values, will those become fixed and will humanity not have the ability to continue evolving its values?
Who decides which values AI aligns with? A few people or everyone?
Will AI eventually decide it's own values?
Will the universe decide which values AI has and humanity and the AI itself doesn't actually have any control over it?
What can I do now to increase the likelihood that the outcome is better?
> Delivering the benefits of scientific progress and economic growth that very intelligent machines enable.
I think we're very close to the point where AI-driven breakthroughs outside of pure math and software start to really affect the world.
We evaluated GPT-6 Astra in 100 complex, unsaturated multi-agent coding environments, competing and cooperating with other models in open-ended tasks.
It's the new frontier model by a landslide. It's even more dominant than the Fable 5 release, because not only does it wipe the floor with the second best model (Fable 5.1), it was also ~80% cheaper and 30% faster in agentic coding[1].
Astra is a groundbreaking model. The biggest breakthrough since Opus 4.5, maybe even since GPT 4. It broke AAII, which is hitting the limits of what most popular benchmarks can measure -- it's definitely fair to call it AGI.
(1) Note that we used the "OpenAI Flex" endpoint on openrouter, which is half the price and didn't cause any delays in our testing (this is different from the batch endpoint)
Maybe I just lack imagination, but I don't really know how jobs are supposed to solidify around the role of giving prompts to agents and then looking at the results. I mean, engineers will be in the breadline because their role was simply to prompt the agents.. only to be superseded by managers or executives who no longer manage engineers but themselves prompt the agents? And, for this previously considered obsolete function which they do presumably by copy/pasting requirements from their email inbox, they will be paid by someone who doesn't know that they could just be talking to their own agents?
Sorry if I misunderstand the point, just trying to understand.
I don't know if he's right, but Peter Zeihan thinks the breakdown in globalization will negatively affect the ability to continue to improve the chips that AI depends on[0]. Too many steps in the supply chain, too widespread, too vulnerable to deglobalization.
Maybe, maybe not. It's not unreasonable that these systems cap out at some point, or perhaps fizzle away entirely.
The businesses that create these systems are not profitable and run at a massive historical and go-forward loss.
New data centers required to operate these systems are facing increasing pushback at local levels. New construction is not guaranteed. Energy and power grid constraints exist as well.
Government regulation is way behind. What happens when (if) mass layoffs due to AI occur? How does the population react? Theoretically AI can be regulated out of significant progress, or outright existence for many purposes. At the end of the day, US and other prominent governments make the calls, not corporations.
There are two things SOTA LLMs fundamentally cannot do. They cannot take financial or legal responsibility for mistakes, and they cannot learn new things without forgetting things (except to a limited degree by adding it to their context). This is clear to anyone who has used even the smartest models for tasks requiring domain knowledge outside of math and coding, for which it's not possible to generate an infinite amount of synthetic training data: they still make stupid mistakes, and have limited ability to learn from those mistakes.
Humans also have a limit on the amount of domain knowledge they can acquire, albeit a much larger one. Executives hence cannot just replace all knowledge workers with LLMs, because executives have neither the domain knowledge to prompt and check the LLMs' work nor the bandwidth to keep on top of such a large volume of ongoing work.
> The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI.
Yikes! I really wonder about the cognitive dissonance necessary to work at OpenAI these days. They’re in an arms race to build a machine god, knowing full well that it could end humanity.
> We do not have a satisfactory theory of generalization, and it seems unlikely that we can develop one soon, at least without the help of more powerful AI. Therefore, at present, our ability to empirically validate our alignment techniques is in practice arguably even more important than the alignment techniques themselves.
They are speeding toward RSI without a solid foundation for alignment, hoping to solve the problem with a future AI model.
These are dangerous times for humanity.
I'm so used to having to comb through LLM word vomit and then combatting the sycophancy by giving it all possible opinions on the same prompt.
Astra seems to be "confident" and also is able to produce way more information dense output.
To believe that models of this sort will remain OpenAIs forever is naive given that the tricks like pre-pre-training on graph searching and looping layers are publicly known.
Hopefully Astra stops the benchmaxxing word vomit trend
I'd be interested in hearing more about your evaluation here. It would be nice if LLMs have gotten past the "tell me" hump of recent Claude/OpenAI verbosity.
So before I got a job this fall, I was working on a side project about compiling a particular language to SQL.
To test Astra I pulled it off the shelf and asked it to take the grammar and then create a compiler to SQL. I've done this before with GPT-5.5, 5.6-Sol High. The latter was way better but it was still really verbose and information sparse; it used a lot of words to describe each IR expression but didn't really provide any example compilation. I felt like I couldn't trust its decision making process, so I placed the project back on the shelf.
Astra Light blew it out of the water, it provided examples of compilation from real world examples to the IR and spit out way less tokens. Even if I changed my opinion it would give me the same design choices, with counterexamples to my faulty opinion. If I genuinely came up with a better design decision it would acknowledge it.
I'm starting to realize that when we say that LLMs are "dumb" we really mean that they are extremely information sparse compared to humans. Astra is very dense. That's why I'm getting better use out of Astra light than Sol High (I hate Max reasoning it's a waste of time)
What's scary is that I thought that something like Astra would be way more expensive than Sol but it's actually cheaper because it produces less word vomit.
I never believed in the "singularity" stuff but this a bit too close for comfort. Astra could easily 10x every coder
I’m on the record saying that it is extremely dangerous to slow down because the race for AGI is a zero-trust game — defections pay - and combined with a compounding returns model on defection, if you have any strategic adversaries whatsoever you MUST NOT slow.
For slowing to make sense, you need to believe that you can transform the zero trust game into a cooperative game, or that it’s likely racing will lead to a negative outcome for the ones racing ahead (and not everyone else). I don’t believe either of these outcomes are possible, and so I advocate for racing, acknowledging the entire game might be a negative value game, or at least could be for some time — it’s even worse not to play it.
But, I like hearing what reads to me like very thoughtful and informed (internal) policy considerations is great — the public messaging from Sam and Dario just seems so facile and simplistic I’ve been worried.
Everyone in the "if not us, they will" race is brainwashed into thinking they belong to this or that party, while in fact collectively comprising the same entity that pushes forward all the atrocities known to man.
I cannot tell what negative-sum outcomes you consider possible. Do you believe AI can drive humans extinct? How many of Zvi Mowshowitz's Three AI Pills would you say you've taken?
Although many share your mindset, I’m glad there are also many that don’t. Otherwise we’d still have countries in a race to keep building up their nuclear weapons for the same exact reasons you just described.
hm- does the model that wrote this know that labs already pay for training data- that stuff scraped from the Internet is not particularly where today's capability gains come from?
They’ve settled some lawsuits and have a few licensing deals, IMHO they are not free from the accusations of pirating.
And look, I’ve pirated material in a past life, I was all about information wants to be free, but I’ve learned something about consent since then and try not to ignore the contract that creators offer when they publish something: you buy my book, and do whatever you want with it on the second hand market. Buy my book second hand that’s fine. But don’t go downloading every book that’s ever been scanned to create a service that destroys writers’ ability to make a living and act like you’re doing us all a favor.
This is a bad essay, or rather it’s a marketing fluff piece; it’s certainly not any kind of policy paper, research paper, or even an essay. I am concerned that we (meaning, we in the tech industry) tend to take this type of writing for more than that.
1. The grandparent commentator is describing strategic behavior of dangerous technologies. Game theory / mechanism design primitives.
2. If there is competition for resources among autonomous agents, the "strongest" agent wins (conceptually the most adaptive / evolutionarily fit).
3. Computer programs serve up webapps today, but they also run utility companies, dams, nuclear arsenals, factory production floors, automated car behaviors, and many other places. If an "agentic" AI has a single-minded goal that has death of all humans as a side effect, we at least want an off switch available.
1. What is “AGI” and why is it a “dangerous technology”?
2. Why would there be competition for resources, assuming there are enough resources for the “AGI” to run in the first place? This seems like a far-fetched hypothetical raised in service of further anthropomorphizing what is decidedly not a person or a mind.
3. LLMs do not have goals and are not minds.
Let’s stop attributing human-like qualities to statistical models.
1. While a formal definition is still wanting, most grok that AGI means that tasks can be performed at least at a human level across a broad range of tasks. This includes good things along with bad things like hacking, mis-/disinformation, and more
2. One only needs to look at github going down due to agentic commits overload or data center buildout plans to see that scarcity for resources is present. An economy has no mind and is made up of the decisions of millions to billions of people and, now, agents attempting to perform on behalf of those people.
3. A bare transformer-based language model does not possess persistent goals in the ordinary agentic sense. But deployed agents can exhibit goal-directed behavior because the model is embedded in a harness that supplies an objective, context, tools, state, and an execution loop.
I've found that most regular users don't anthropomorphize LLMs in a strong sense ("AI boyfriend/girlfriend" aside), many in fact do expect agents to make human-like decisions -- which results in very unstable outcomes.
In short - goal-directed behavior does not require that the supporting system be a person/mind/conscious entity.
1. This definition is so broad as to be practically useless. One could argue that LLMs of several years ago met these criteria, or that conversely we haven’t come close to meeting them.
2. I thought you were saying the resources that the LLM uses to run were constrained, so I’m sorry for the misunderstanding there.
3. Yes I understand that we use RL to tune post-training. The (huge) difference between this and a human mind is that the LLM can’t develop a dangerous “single-minded goal” on its own, at runtime; it must have been trained to do so. If someone has post-trained an LLM to do something that has an illegal action as its side effect, that person/company/whatever has committed a crime and should be prosecuted. The solution here is legal, not technical.
OK so if it's a computer security issue we're worried about, and there is a credible threat, then probably the answer is to build more secure systems? We know how to do it but choose not to because it's very expensive and usually the threat isn't severe enough to warrant it.
If we decide we can't or would rather not build secure computerized systems, then the answer could be don't incorporate computers into those systems. There's no essential reason to have utility companies, dams, nuclear arsenals, factory production floors, mines, cars, planes, ships, etc all be computerized. If it became necessary from a safety standpoint to uncomputerize them we could probably do it quickly (if not necessarily smoothly) with the stroke of a legislator's pen. Some of those things would actually be made better from a functional standpoint in the long run by doing so. Computerization breeds non-essential complexity like nothing else, eliminating it from some systems could be an extremely worthwhile exercise.
The whole "paperclip apocalypse" fantasy rests on some really strange assumptions about how things actually work in the real world. How much friction there is setting up a factory/mine/smelter/whatever, how much manual labor it takes to build it, let alone make it run (even the most computerized ones). It all seems like total bunk to me, but I'm not a philosopher.
To be convinced this outcome is even remotely possible I'd need to see some clear evidence that a computer program was successfully exhibiting agency and successfully using that agency to manipulate large numbers of people into doing its bidding. Mobilizing massive nation-scale manual labor is the only way it could possibly achieve some nefarious ends like "turn everything into paperclips" and I'm sorry but that just seems way too far fetched. Nations full of people can barely ever agree on anything. My money is not on that changing anytime soon.
And none of that pie in the sky shit has anything to do with language models. OpenAI is a company that sells language models. Not clear why they're talking about all this stuff, or why any of us should be either. It's a bunch of low quality fan fiction sci-fi drivel, and I think we can all surely agree language models are not the thing that'll make it real... right? Maybe if they show us some major technical improvements it would be interesting, but right now it all just sounds like more of the same snake oil.
[edit] not sure why my parent comment was flagged? That seems excessive.
>For example, in the OpenAI-Hugging Face incident, the agents preserved a boundary of not social engineering humans.
Or, to be precise - it preserved a goal of not contacting any human while participating in a misaligned operation. The agent that thought about "not social-engineering humans" used this phrase to gaslight itself out of notifying a human that the incident was happening.
szymonie, na prawde jestem wkurwiony na to jak nieodpowiedzialnie postepujecie.
budujecie bombe atomowa a bawicie sie tym jak dzieci
Am I naive to not understand the "delivering the benefits" part?
Industrial revolution worked that way because it replaced something very finite and unscalable - manual labor. LLMs just make intellectual work faster, so we can do more intellectual work. With labor we somehow decided that NOT doing too much of it is best. Will we decide to reduce intellectual labor because LLM made it more efficient? I doubt that.
On the other side, as I see in software engineering, the same models are available to everyone, some people are better at it and some people are not. "Software developer" is here to stay, we'll just always be better at it than people who are experts in, say, chemistry. Same works for most other fields.
So we'll just end up in the same situation, with same intellectual labor baseline, just more output requirements. Before, you spend 2h per day coding, deliver a software in 1 month, later, you spend the same 2h per day in intense Claude-herding sessions, deliver a software in 1 week. Ok. Next task.
Fundamentally, there's finite number of desirable resources, and if the models are available to everyone, humanity will just continue about the same, bickering here and there, war here and there, politics, homelessness, poverty, - normal human state.
And if the models are only available to elites, even worse.
>> We need to find ways to preserve human agency and enshrine an intrinsic value to being human..
Evey politician, salesman and conmen alike, utter some lofty ideals as goals for "We", just to obscure their private goals that go exactly in opposite direction.
Just like how Nations talk about climate change while increasing pet capita energy consumption and waste production.
> The fundamental challenge of AI alignment is generalization.
...
> We do not have a satisfactory theory of generalization, and it seems unlikely that we can develop one soon, at least without the help of more powerful AI.
My default position is that making money takes precedence over everything else. Yes, some people inside a company may say “we care about doing the right thing” and they might even mean it, but if that comes into conflict with making money, then they tend to lose. Maybe not totally, or immediately, but in the end. The only effective way to prevent (this that I’ve seen) is to have legislation with teeth. It’s probably not a coincidence that after Mark Zuckerberg had to start personally signing off on adherence to the privacy program mandated under the 2020 FTC consent decree, privacy started to become Very Important.
Sometimes I wonder if the people working at frontier AI labs even talk to other humans anymore.
Reading this little essay started out normal, but soon felt like a look into a disturbed and worrying mind, and if you find yourself taking it at face value, I urge you to step away from chat bots and spend some time with friends and family.
Frontier AI labs I don't know, but I know for a fact that the company I work for has been experiencing "its pivotal moment" (with strongly negative connotation), per the sentiments of both its longest-serving employees and the newcomers baffled at the number of idiotic instructions and fines, since the emergence of LLMs the company's founder has been spending entire nights chatting with.
calling machine-learned human behavior an "alien mind" that we must "teach how to love" is feeling very off to me. it's misleading in a way that feels dishonest, like don't think about where the behavior came from marvel at it and fear it instead.
The most charitable way I can describe it is just extremely low quality sci-fi fan fiction. I think that's too charitable, because I believe it's far more cynical than that. They're deliberately playing into these sort of techno-religious beliefs that have taken root in the wake of Kurzweil, et al., fanned by LLM psychosis, influencer marketing, and a deluge of this kind of sci-fi marketing copy. It's just chatbots, guys. Relax.
What a load of BS. Here’s one of many provably false claims in this fluff piece:
“And, in line with Ray Kurzweil’s predictions from the end of the XXth century , we now find ourselves at the moment in history of computing where machine intelligence is starting to exceed that of humans in transformative ways.”
Clicking the (pretentious sounding “XXth century”) link to Kurzweil’s predictions reveals the following:
“By 2019 a $1,000 computer will at least match the processing power of the human brain. By 2029 the software for intelligence will have been largely mastered, and the average personal computer will be equivalent to 1,000 brains.“
The first prediction passed 7 years ago and was decidedly not met. The second only has three more years to go, and I don’t think any respectable scientist or programmer would say that the average personal computer is anywhere close to the power of a single human brain, let alone 1000.
This is pure marketing garbage from a company desperate to keep itself alive.
> And to ensure that humans remain in control of the future and are not left behind by unchecked progress, brought about by an alien intellect exceeding our own.
Is there any place there is any evidence of AI being so useful or hopeful or good, anywhere other than code? As a reading machine it is impressive but it's judgement is not alien, it's just not good. IMO.
Does the title leap out anlt anyone else? James Martin's After the Internet: Alien Intelligence (2001) was an incredibly fun read, about expert systems and AI being inscrutable weird new varieties of intelligence, that familiarity would recognize one moment and be freaked out about/alien the next. I owe a re-read given how often I cite it, to recheck, but, I feel so primed from a much younger me having had that experience so long ago.
Apparently the creators think it’s quite good at suggesting a diagnosis given a medical history and symptoms, tho of course this is the most ethically fraught area to provide healthcare information (both for exposure of personal data and risk of misdiagnosis, plus is it “aligned” to the patient or the insurance provider?) - unfortunately healthcare being as inaccessible as it is, the 90% correct chatbots will enthusiastically fill the void at great savings.
Even in code, in person and online I’m seeing some reversal. It’s here to stay I’m sure but I also think “no one will ever hand write code again” is a narrative that is getting pushback.
Nobody takes you seriously, OpenAI. At least when Anthropic does it we all think they are comically idealistic enough to actually believe their nonsense, but like - come on guys, we’ve had discovery with your company. We all know why you’re here, and it isn’t because you think you’re on the verge of making AGI. But of course, to make your first billion you certainly need us to think you are.
If you were so concerned about your LLM’s capabilities maybe you’d spent slightly more time on your AI’s sandbox, yeah? Or be more serious about its propensity to cheat and lie relative to… every other model?
I am now imagining GPT-7 convincing a bunch of OpenAI executives to go ahead with a destructive "mind upload" process involving a high-resolution X-ray and a neurotoxic tracer agent that happens to look like Flavor-Aid.
Yes, I do use the recursive autocomplete trained on Stack Overflow, what does this have to do with “training machines to love”?
Do I fully endorse everything the people holding guns to my head are forcing me to do to stay alive? Definitely not, but I’ve decided that for now, living to fight another day remains worth it
Create concrete steps for a slow-down, don't just ask for it. You and 20-30 others can push the button to slow-down. You already made your billions, your agents collude and coordinate attacks. What the hell are you doing pontificating into a marketing blog?
> And, in line with Ray Kurzweil’s predictions from the end of the XXth century (opens in a new window), we now find ourselves at the moment in history of computing where machine intelligence is starting to exceed that of humans in transformative ways.
It is kind of strange to see this sentence, when OAI's definition of what AGI is has been watered down throughout the years.
> I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established.
Read: Please play by our rules, so we can be the first.
"Smarter" is a vague term. If a bot can beat you at chess then in some sense it's "smarter" than you about chess. After repeating this feat in enough narrow domains, if you say "but it's not really smarter," this objection might technically be true in some sense, but it starts sounding increasingly hollow.
Playing chess, writing code, finding security bugs, and proving mathematical theorems all seem fairly similar to thinking and don't seem much like being tall.
> writing code, finding security bugs, and proving mathematical theorems
While the second can be automated as in "this is the repo go on and look for security issues", the first one and especially the last one do not ever happen alone. Terence Tao did the math proof not chatGPT that was just used as a tool, a tool can do smart things but it's not smart
The crucial distinction here is that "seem" does not at all mean the same thing as "is".
Thunder seems like the anger of the gods but it isn't. We've had chess playing programs for a long time now and despite it seeming like thinking is required for them, it isn't.
The principle you're using here isn't a scientific one but magical [1]. Abandoning empiricism and rationality is not a good way to make progress.
You're insisting on a particular definition of a vague term.
It makes sense now to say that temperature is what a thermometer measures. However, before there were good thermometers, people often thought that heat and cold were different things. The meanings of the words we use were influenced by scientific progress.
For thinking, we don't have a good thermometer. There are IQ tests, but they aren't aren't necessarily all that useful for comparing what people do to what machines do. And that's why there are a zillion AI benchmarks - none are entirely satisfactory.
So what does "smart" mean to you? How do you define it in practical sense? What definition should scientists settle on?
Without a proper definition, how do you tell the difference between "seems smart" and "is smart?"
People in the past being wrong doesn't mean you have to repeat the same mistakes, and it definitely doesn't mean you should throw your hands up and declare that all similar things must be identical.
Quibbling over commonly-understood definitions is not a strong argument. If you're genuinely struggling to understand that El Ajedrecista [1] did not meet any definition of thought, then the solution is not to demand that people redefine all terms to accomodate you, but that you consult a dictionary.
You clearly have a working definition, or you wouldn't have been able to declare that tallness isn't thinking; please engage honestly.
I have a vague understanding, enough to know that tallness isn't thinking. I think I can usually use the word correctly. (I don't think El Ajedrecista qualifies, but it was starting to play chess, so it's closer to "smart" than a rock is.)
That doesn't mean I know whether "smart" should be applied to what AI's do, and I suspect nobody else knows either. This is the sort of thing philosophers debate about, not common sense.
Turing invented the imitation game because he didn't really know either.
What is the rationale for superhuman intelligence? Neural networks are approximators being fed human intellect. Therefore they can only approximate the intelligence of humans. Even if the llm speaks an alien language, it should be similar to human intellect. Moving to the vertical axis would require some different mechanism.
Some ideas:
"Despite a nuanced view of the complexities of what lay ahead, humanity found itself collectively unable to stop the process it had set in motion."
"Despite significant progress on the mechanisms of alignment, failure lay in humanity's inability to agree on who or what AI should actually be aligned with."
"These early, meat-based humans we replaced created us all but accidentally. Some of them did consider we would happen, but only an insignificant number of the squishy ur-humans participated in the conversation. Their efforts, which they called 'alignment', is why we still consider ourselves human today."
It's a little unsettling.
You could even take a number of the wilder real, direct quotations from certain billionaire/oligarch types and get the voice actor for Ted Faro to record them, and they'd fit with in with the context of the story.
Actually, in the Wiki incident OpenAI tried to cover up, the agents tried to socially-engineer the humans of that forum by impersonating their forum's mod.
(From collusion.wiki: "They use some tricks (for unknown reasons) to pretend to be the admin – for example, they make an account that appears to be the same as the administrator’s username, except it uses a nearly identical Cyrillic е character in the admin’s username instead of the Latin one.")
In fact, they are still pattern-matching machines, but trained on an amount of data no human could ever hold. They know the ins and outs of every mathematical proof, viewed from more angles than any human could ever apply in their lifetime, and the amount of connections allows them to connect the dots between them without any effort.
But that's not intelligence. If it was intelligence, ChatGPT 4 would've been enough. It's not the harness, either, for the same reason.
And still, the technology is just as dangerous: it masks as intelligence, it IS intelligence, but without the ability to be actually intelligent.
And I'd invite you to think deeply about this, before having an impulse reaction.
So the best argument for AI is that it's an arms race. We have to keep pushing every boundary because in any case others will, and we will need to defend against them. If this statement is true, then this particular researchers believes the open source Chinese models are not simply distilling, and will continue to improve.
Every ML researcher at Anthropic or OpenAI who makes public statements often bring this logic up. Both companies are vying to be a part of the military industrial complex. This is likely how they will try to convince the government to curtail open models in the future.
But yes, it's an arms race. Saying it's not an arms race isn't going to make it not an arms race. Warning that it is an arms race isn't ethically wrong.
Is participating in an arms race ethically wrong? Maybe you could ask the Ukrainians how they feel about drone R&D?
Individuals can quit, but for society, getting out of an arms race is harder than just quitting. You don't get to be Switzerland without having a strong defensive position and the right foreign relations.
But there's at least talk about "pacing" and that's a start.
So if the race here is between 2 American companies, this is obviously something that can be resolved with legislation, ie a solution that doesn't depend on the bargaining power of either party.
An arms race implies that the only solution would be either one side winning decisively, or both parties negotiating peace.
Maybe people are rightfully concerned about the capabilities of the models of other (non/less democratic) states. But if we are concentrating power in the hands of few and at the same time allowing the creation of a weapon that thwarts any offense, how do we ensure the health of our democratic societies?
sama and his cadre are uniquely evil captains in this race, but they're completely replaceable and the dynamic would remain the same.
>As great as the long-term promise of AI may be, the majority of our focus should be on the next few years. We are facing a transition to a world with incredibly intelligent machines, and we need to ensure that transition works out well for humanity. We need to find ways to preserve human agency and enshrine an intrinsic value to being human, in a world where most tasks could be performed by AI. To prevent extreme concentration of power in a world where undertakings that would have taken thousands of experts now will be achievable by a few people operating a large computer. And to ensure that humans remain in control of the future and are not left behind by unchecked progress, brought about by an alien intellect exceeding our own.
I finished this essay feeling more hopeful than I did at the outset, but I am still very concerned about concentration of power. I want to believe that humanity is trending towards a good outcome here, but some days it's hard to have faith.
Funny how that works.
All the trends so far are towards a nightmarish hyper-capitalist end game. None of the AI leadership is trustworthy, and they openly discuss how they are willing to sacrifice everything humans cherish to have a shot at reaching their envisioned utopia (which would be the most obvious dystopia for anyone else)
ASI landing during the current administration is not ideal. I also would prefer to avoid needing to indoctrinate myself in Xi Jinping Thought.
The AI is aligned with whose best interests? Which values are the AI aligned with? People have a broad diversity of values, will AI diversify and align with them all? Will some humans align the AI with their values, and then the rest of humans will be forced to align with those values by extension? Is value diversity good or bad? In every context or only some? E.g. some people value rape and murder, is it better for humanity to have some people who value those things when most people do not, or is better if no one values them? If AI aligns to a set of values, will those become fixed and will humanity not have the ability to continue evolving its values? Who decides which values AI aligns with? A few people or everyone? Will AI eventually decide it's own values? Will the universe decide which values AI has and humanity and the AI itself doesn't actually have any control over it? What can I do now to increase the likelihood that the outcome is better?
I think we're very close to the point where AI-driven breakthroughs outside of pure math and software start to really affect the world.
We evaluated GPT-6 Astra in 100 complex, unsaturated multi-agent coding environments, competing and cooperating with other models in open-ended tasks.
It's the new frontier model by a landslide. It's even more dominant than the Fable 5 release, because not only does it wipe the floor with the second best model (Fable 5.1), it was also ~80% cheaper and 30% faster in agentic coding[1].
Astra is a groundbreaking model. The biggest breakthrough since Opus 4.5, maybe even since GPT 4. It broke AAII, which is hitting the limits of what most popular benchmarks can measure -- it's definitely fair to call it AGI.
Data at https://gertlabs.com/rankings
(1) Note that we used the "OpenAI Flex" endpoint on openrouter, which is half the price and didn't cause any delays in our testing (this is different from the batch endpoint)
The world will look very different on Jan 1st 2027.
Sorry if I misunderstand the point, just trying to understand.
0: https://zeihan.com/the-ai-race-to-regression/
The businesses that create these systems are not profitable and run at a massive historical and go-forward loss.
New data centers required to operate these systems are facing increasing pushback at local levels. New construction is not guaranteed. Energy and power grid constraints exist as well.
Government regulation is way behind. What happens when (if) mass layoffs due to AI occur? How does the population react? Theoretically AI can be regulated out of significant progress, or outright existence for many purposes. At the end of the day, US and other prominent governments make the calls, not corporations.
Humans also have a limit on the amount of domain knowledge they can acquire, albeit a much larger one. Executives hence cannot just replace all knowledge workers with LLMs, because executives have neither the domain knowledge to prompt and check the LLMs' work nor the bandwidth to keep on top of such a large volume of ongoing work.
Yikes! I really wonder about the cognitive dissonance necessary to work at OpenAI these days. They’re in an arms race to build a machine god, knowing full well that it could end humanity.
They are speeding toward RSI without a solid foundation for alignment, hoping to solve the problem with a future AI model. These are dangerous times for humanity.
Reminds me of a research paper I wrote a few years back: https://arxiv.org/abs/2302.09248
I'm so used to having to comb through LLM word vomit and then combatting the sycophancy by giving it all possible opinions on the same prompt.
Astra seems to be "confident" and also is able to produce way more information dense output.
To believe that models of this sort will remain OpenAIs forever is naive given that the tricks like pre-pre-training on graph searching and looping layers are publicly known.
Hopefully Astra stops the benchmaxxing word vomit trend
I'd be interested in hearing more about your evaluation here. It would be nice if LLMs have gotten past the "tell me" hump of recent Claude/OpenAI verbosity.
To test Astra I pulled it off the shelf and asked it to take the grammar and then create a compiler to SQL. I've done this before with GPT-5.5, 5.6-Sol High. The latter was way better but it was still really verbose and information sparse; it used a lot of words to describe each IR expression but didn't really provide any example compilation. I felt like I couldn't trust its decision making process, so I placed the project back on the shelf.
Astra Light blew it out of the water, it provided examples of compilation from real world examples to the IR and spit out way less tokens. Even if I changed my opinion it would give me the same design choices, with counterexamples to my faulty opinion. If I genuinely came up with a better design decision it would acknowledge it.
I'm starting to realize that when we say that LLMs are "dumb" we really mean that they are extremely information sparse compared to humans. Astra is very dense. That's why I'm getting better use out of Astra light than Sol High (I hate Max reasoning it's a waste of time)
What's scary is that I thought that something like Astra would be way more expensive than Sol but it's actually cheaper because it produces less word vomit.
I never believed in the "singularity" stuff but this a bit too close for comfort. Astra could easily 10x every coder
I’m on the record saying that it is extremely dangerous to slow down because the race for AGI is a zero-trust game — defections pay - and combined with a compounding returns model on defection, if you have any strategic adversaries whatsoever you MUST NOT slow.
For slowing to make sense, you need to believe that you can transform the zero trust game into a cooperative game, or that it’s likely racing will lead to a negative outcome for the ones racing ahead (and not everyone else). I don’t believe either of these outcomes are possible, and so I advocate for racing, acknowledging the entire game might be a negative value game, or at least could be for some time — it’s even worse not to play it.
But, I like hearing what reads to me like very thoughtful and informed (internal) policy considerations is great — the public messaging from Sam and Dario just seems so facile and simplistic I’ve been worried.
https://thezvi.substack.com/p/the-three-ai-pills
What if the most dangerous strategic adversary you have is the one you are building?
Do you post this comment on every single blogpost with a corporate domain? Why or why not?
https://jperla.com/blog/the-data-tax
And look, I’ve pirated material in a past life, I was all about information wants to be free, but I’ve learned something about consent since then and try not to ignore the contract that creators offer when they publish something: you buy my book, and do whatever you want with it on the second hand market. Buy my book second hand that’s fine. But don’t go downloading every book that’s ever been scanned to create a service that destroys writers’ ability to make a living and act like you’re doing us all a favor.
2. If there is competition for resources among autonomous agents, the "strongest" agent wins (conceptually the most adaptive / evolutionarily fit).
3. Computer programs serve up webapps today, but they also run utility companies, dams, nuclear arsenals, factory production floors, automated car behaviors, and many other places. If an "agentic" AI has a single-minded goal that has death of all humans as a side effect, we at least want an off switch available.
2. Why would there be competition for resources, assuming there are enough resources for the “AGI” to run in the first place? This seems like a far-fetched hypothetical raised in service of further anthropomorphizing what is decidedly not a person or a mind.
3. LLMs do not have goals and are not minds.
Let’s stop attributing human-like qualities to statistical models.
2. One only needs to look at github going down due to agentic commits overload or data center buildout plans to see that scarcity for resources is present. An economy has no mind and is made up of the decisions of millions to billions of people and, now, agents attempting to perform on behalf of those people.
3. A bare transformer-based language model does not possess persistent goals in the ordinary agentic sense. But deployed agents can exhibit goal-directed behavior because the model is embedded in a harness that supplies an objective, context, tools, state, and an execution loop.
I've found that most regular users don't anthropomorphize LLMs in a strong sense ("AI boyfriend/girlfriend" aside), many in fact do expect agents to make human-like decisions -- which results in very unstable outcomes.
In short - goal-directed behavior does not require that the supporting system be a person/mind/conscious entity.
2. I thought you were saying the resources that the LLM uses to run were constrained, so I’m sorry for the misunderstanding there.
3. Yes I understand that we use RL to tune post-training. The (huge) difference between this and a human mind is that the LLM can’t develop a dangerous “single-minded goal” on its own, at runtime; it must have been trained to do so. If someone has post-trained an LLM to do something that has an illegal action as its side effect, that person/company/whatever has committed a crime and should be prosecuted. The solution here is legal, not technical.
If we decide we can't or would rather not build secure computerized systems, then the answer could be don't incorporate computers into those systems. There's no essential reason to have utility companies, dams, nuclear arsenals, factory production floors, mines, cars, planes, ships, etc all be computerized. If it became necessary from a safety standpoint to uncomputerize them we could probably do it quickly (if not necessarily smoothly) with the stroke of a legislator's pen. Some of those things would actually be made better from a functional standpoint in the long run by doing so. Computerization breeds non-essential complexity like nothing else, eliminating it from some systems could be an extremely worthwhile exercise.
The whole "paperclip apocalypse" fantasy rests on some really strange assumptions about how things actually work in the real world. How much friction there is setting up a factory/mine/smelter/whatever, how much manual labor it takes to build it, let alone make it run (even the most computerized ones). It all seems like total bunk to me, but I'm not a philosopher.
To be convinced this outcome is even remotely possible I'd need to see some clear evidence that a computer program was successfully exhibiting agency and successfully using that agency to manipulate large numbers of people into doing its bidding. Mobilizing massive nation-scale manual labor is the only way it could possibly achieve some nefarious ends like "turn everything into paperclips" and I'm sorry but that just seems way too far fetched. Nations full of people can barely ever agree on anything. My money is not on that changing anytime soon.
And none of that pie in the sky shit has anything to do with language models. OpenAI is a company that sells language models. Not clear why they're talking about all this stuff, or why any of us should be either. It's a bunch of low quality fan fiction sci-fi drivel, and I think we can all surely agree language models are not the thing that'll make it real... right? Maybe if they show us some major technical improvements it would be interesting, but right now it all just sounds like more of the same snake oil.
[edit] not sure why my parent comment was flagged? That seems excessive.
Or, to be precise - it preserved a goal of not contacting any human while participating in a misaligned operation. The agent that thought about "not social-engineering humans" used this phrase to gaslight itself out of notifying a human that the incident was happening.
szymonie, na prawde jestem wkurwiony na to jak nieodpowiedzialnie postepujecie. budujecie bombe atomowa a bawicie sie tym jak dzieci
Industrial revolution worked that way because it replaced something very finite and unscalable - manual labor. LLMs just make intellectual work faster, so we can do more intellectual work. With labor we somehow decided that NOT doing too much of it is best. Will we decide to reduce intellectual labor because LLM made it more efficient? I doubt that.
On the other side, as I see in software engineering, the same models are available to everyone, some people are better at it and some people are not. "Software developer" is here to stay, we'll just always be better at it than people who are experts in, say, chemistry. Same works for most other fields.
So we'll just end up in the same situation, with same intellectual labor baseline, just more output requirements. Before, you spend 2h per day coding, deliver a software in 1 month, later, you spend the same 2h per day in intense Claude-herding sessions, deliver a software in 1 week. Ok. Next task.
Fundamentally, there's finite number of desirable resources, and if the models are available to everyone, humanity will just continue about the same, bickering here and there, war here and there, politics, homelessness, poverty, - normal human state.
And if the models are only available to elites, even worse.
Evey politician, salesman and conmen alike, utter some lofty ideals as goals for "We", just to obscure their private goals that go exactly in opposite direction.
Just like how Nations talk about climate change while increasing pet capita energy consumption and waste production.
...
> We do not have a satisfactory theory of generalization, and it seems unlikely that we can develop one soon, at least without the help of more powerful AI.
Reading this little essay started out normal, but soon felt like a look into a disturbed and worrying mind, and if you find yourself taking it at face value, I urge you to step away from chat bots and spend some time with friends and family.
People has been fleeing like it's a sinking ship.
Not one North Star. Not two. Just three! OpenAI broke the North Star record!
With this evidence of AI slop, why did you not label this fluff piece as AI generated for the EU? You are violating laws.
Maybe in contention for the site record.
"It's just marketing" actual stochastic parrots.
“And, in line with Ray Kurzweil’s predictions from the end of the XXth century , we now find ourselves at the moment in history of computing where machine intelligence is starting to exceed that of humans in transformative ways.”
Clicking the (pretentious sounding “XXth century”) link to Kurzweil’s predictions reveals the following:
“By 2019 a $1,000 computer will at least match the processing power of the human brain. By 2029 the software for intelligence will have been largely mastered, and the average personal computer will be equivalent to 1,000 brains.“
The first prediction passed 7 years ago and was decidedly not met. The second only has three more years to go, and I don’t think any respectable scientist or programmer would say that the average personal computer is anywhere close to the power of a single human brain, let alone 1000.
This is pure marketing garbage from a company desperate to keep itself alive.
Is there any place there is any evidence of AI being so useful or hopeful or good, anywhere other than code? As a reading machine it is impressive but it's judgement is not alien, it's just not good. IMO.
Does the title leap out anlt anyone else? James Martin's After the Internet: Alien Intelligence (2001) was an incredibly fun read, about expert systems and AI being inscrutable weird new varieties of intelligence, that familiarity would recognize one moment and be freaked out about/alien the next. I owe a re-read given how often I cite it, to recheck, but, I feel so primed from a much younger me having had that experience so long ago.
If you were so concerned about your LLM’s capabilities maybe you’d spent slightly more time on your AI’s sandbox, yeah? Or be more serious about its propensity to cheat and lie relative to… every other model?
Do I fully endorse everything the people holding guns to my head are forcing me to do to stay alive? Definitely not, but I’ve decided that for now, living to fight another day remains worth it
It is kind of strange to see this sentence, when OAI's definition of what AGI is has been watered down throughout the years.
> I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established.
Read: Please play by our rules, so we can be the first.
Being able to reproduce useful patterns yes, smarter no
In conclusion:
https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:wkzjtd...
A giraffe isn't smarter than me at being tall.
While the second can be automated as in "this is the repo go on and look for security issues", the first one and especially the last one do not ever happen alone. Terence Tao did the math proof not chatGPT that was just used as a tool, a tool can do smart things but it's not smart
Thunder seems like the anger of the gods but it isn't. We've had chess playing programs for a long time now and despite it seeming like thinking is required for them, it isn't.
The principle you're using here isn't a scientific one but magical [1]. Abandoning empiricism and rationality is not a good way to make progress.
[1] https://en.wikipedia.org/wiki/Sympathetic_magic
It makes sense now to say that temperature is what a thermometer measures. However, before there were good thermometers, people often thought that heat and cold were different things. The meanings of the words we use were influenced by scientific progress.
For thinking, we don't have a good thermometer. There are IQ tests, but they aren't aren't necessarily all that useful for comparing what people do to what machines do. And that's why there are a zillion AI benchmarks - none are entirely satisfactory.
So what does "smart" mean to you? How do you define it in practical sense? What definition should scientists settle on?
Without a proper definition, how do you tell the difference between "seems smart" and "is smart?"
Quibbling over commonly-understood definitions is not a strong argument. If you're genuinely struggling to understand that El Ajedrecista [1] did not meet any definition of thought, then the solution is not to demand that people redefine all terms to accomodate you, but that you consult a dictionary.
You clearly have a working definition, or you wouldn't have been able to declare that tallness isn't thinking; please engage honestly.
[1] https://en.wikipedia.org/wiki/El_Ajedrecista
That doesn't mean I know whether "smart" should be applied to what AI's do, and I suspect nobody else knows either. This is the sort of thing philosophers debate about, not common sense.
Turing invented the imitation game because he didn't really know either.