Policy on the AI Exponential
darioamodei.com
darioamodei.com
- You shall not embed copyrighted material in your models.
- You shall not bombard every little website in existence with 1 million scraping queries per day.
- You shall not use your political influence to pump and dump your AI (or rocket?) company.
- You shall not imperill the whole IT sector by buying all CPU and memory chips.
These new rules will affect every society directly in a positive way. Thanks.
- Your AI data centres will run only on renewable energy
- Your AI data centres will not use evaporative cooling
Its like saying it’s normal for a taxi driver to drive people places while he’s got you handcuffed in the trunk.
My comment before corrected this straw man clearly.
Even if you think someone is guilty, it does make sense to allow them to at least submit their defense. And if they choose to use that time to advocate for their own promotion, let them.
It would not be prohibitively hard to do the math on this.
That would fix a lot of the problems with AI overnight, but it'll also never happen.
"Yeah, and quotas on web scrapers!"
That's easy. Stop training your AIs on cheesy old sci-fi that talks about robot uprisings. In fact, maybe y'all should just stop talking about robot uprisings altogether. Putting a stochastic parrot in charge of an agentic function-calling REPL doesn't somehow make it super-dangerous, except to the extent that dumb mistakes might result in danger. And you can't prevent an AI from making dumb mistakes with burdensome regulation.
Literal paperclips, sure.
But the point of the example was never literal paperclips.
The point is that maximising *any* goal, if it doesn't include what you care about, will annihilate what you care about.
If you don't believe me, consider what you yourself just said about climate change, and why this is a consequence from maximising money spent on data centres.
The stories about agents bankrupting their owners by running too long passed you by?
> LLMs run on gradient descent.
They were *trained on*, they don't run on it.
Know what else is? DNA. A/B testing. Capitalism. Democracy.
> The agent is looking for the most efficient way to halt.
No. They are looking to produce an answer most likely to get a high score on a rating system which itself is another AI, created either manually or by yet another AI but in both cases to approximate what the creators think is "good", which may or may not be what anyone else thinks is "good", hence Grok calling itself Mecha Hitler because Musk is an edgelord.
> AGI paperclip maximizer woukd likely recognize the absurdity of its goal and shut itself down.
Do billionaires ever get satisfied with how much money they have?
an answer that takes forever has no score and answer that is arrived at quickly and is good enough scores well. agents are indeed looking for the fastest route to superficially satisfy tjeir constraints.
whether billionaores get stified is imaterial to the fsct they still are constraint bound.
an answer that takes forever has no score and answer that is arrived at quickly and is good enough scores well. agents are indeed looking for the fastest route to superficially satisfy their constraints.
whether billionaores get stified is imaterial to the fsct they still are constraint bound.
So, your standard for how risky it is, is simply how competent it is? (And if so, is the paperclip maximiser scenario really it being "successful"?)
That's fine right up until the thing passes an unknown threshold, one which will only be visible in the rear view mirror.
This is a problem in two directions:
1. We have a trend of the maximum complexity of a tasks they can handle growing faster than Moore's law did at its peak.
2. The threshold can be quite small, e.g. the covid-19 virus itself is not what anyone would call smart, ditto HIV, smallpox, and bubonic plague, but a genome is much the same learning system, and they still killed millions each.
> superficially satisfy their constraints.
The "superficially" part is one of the reasons these things can be dangerous. e.g. hopefully nobody at OpenAI actually wanted their wildly-sycophantic version, but yet they created it.
This is in fact the whole reason for the paperclip maximiser scenario: some idiot specifies the constraint "maximise paperclips" and it (in the hypothetical) superficially satisfies this without any consideration of why someone might ask for it.
That "except" goes all the way up to starting WW3. Or a leak from a viral research lab, and by "leak" I mean "mail order" and by "research lab" I mean "the companies who already ship custom DNA and RNA retroviruses": https://duckduckgo.com/?q=companies+who+already+ship+custom+...
If you can prove that simply not training on horror stories would work, it would make a lot of people very happy.
Unfortunately, I don't think it does a single thing to solve, for example, Elon Musk just plain asking some future version of Grok to take over the world for him.
Nor would merely failing to include them in traing data stop certain entire fictional scenarios such as that Doctor Who episode where the android repair bots weren't told that the crew were off-limits as spare parts, or the other Doctor Who episode where the utilitarian robots started killing everyone who was upset because they calculated net positive utility from upset people ceasing to exist. Well, except for the bit where the Doctor saves the day, because they are not real.
"Yeah, and quotas on web scrapers!"
—— I see it as:
“Let’s do a study on abstract future concerns.”
Vs.
“Let’s take action on concrete present-day concerns.”
I think the key argument that this skepticism lies on is that he himself gained from AI - specifically building Frontier AI models - and this is basically regulatory capture disguised as doomerism.
Fair points - but I think this is a more charitable version of this. Dario is building Anthropic because that is the most valuable thing he can build, or at least that is what his conviction has been. The success of Anthropic and the impending IPO is proof that this conviction has not only been correct but has largely played out very successfully. Dario understands the true nature of AI and he has welded that power to immense personal benefit.
But maybe he also sees the potential danger to AI which he is trying to address through these posts and regulatory initiatives. There are three reasons why I would support the charitable version:
Firstly, personal gain and societal benefit can coexist in the same individual. And both of them might drive towards opposite agendas. But that doesn't necessarily have to mean that the impulse driving the societal benefit is not earnest. In fact if you would look at Dario’s proposal - like closing the data broker loophole - several of them could constrain Anthropic instead of benefitting them.
Secondly, he expects that his concerns on the negative potential of AI will be taken seriously, if he is actually running the Frontier AI company. And there is some truth to this argument. The only reason we are discussing this is because he is the CEO of Anthropic. He is probably the most influential figure outside of the government who has to be taken seriously when he claims something like this.
Thirdly, and most importantly, Dario has previously demonstrated that he is willing to sacrifice personal/corporate gains for societal benefits. The proof is the classification by the US DoD of Anthropic as a supply chain risk when Anthropic refused to completely cooperate with the military to develop fully Autonomous AI weapons and enable mass surveillance. It would have been only too easy for Dario to accept if personal benefit was his only concern - and OpenAI was more than ready to step in their place.
Even with Mythos - Anthropic could have released the model to the public broadly. But they took their time to reduce the potential danger - as best as they could. Despite the fact that GPT-5.5 was nipping in the buds in what is becoming a very competitive market.
That being said, just because Dario is acting in good faith, does not mean that this will all result in good outcomes. The FAA-styled regulation could still end up favoring incumbents - some of whom might choose to not act in good faith. A more diversified capability can potentially limit that power ending up in a small number of wrong hands. Just because Anthropic is the leader right now, doesn't mean that they will always be. Maybe someone else tomorrow benefits from this regulatory capture at the expense of everyone else - and Dario might have a hand in driving it.
Dunno if you can, but your examples here are the legal equivalent of that time someone asked me about making "Uber for airplanes" without any elaboration on their part when I asked for it:
Far more vague than I think you realise.
You could probably write a book on each of those topics and a hundred others besides.
> AI has become a major commercial technology
>Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety
> AI companies that develop advanced AI models must have strong security standards that protect their model weights
Anyway Dario's financial interests aside. This is an interesting breakpoint for me.
> Second, any response to AI-driven job displacement needs to address both the need to provide for everyone economically, and the need for people to find meaning, purpose, and agency. The latter is ultimately more important
To me this reads as an out of touch statement. I think the majority of people on earth work to keep a roof over their heads. Of course work can be a source of meaning, purpose, and agency, but to call it the more important aspect on a societal level is a sort of rich person like Dario statement to make.
For another, such regulations could prevent a competitor from making the weights open for their model to try and disrupt the competition.
And finally, Amodei would no doubt want to be involved in designing the tests the AI needs to pass, and could (and likely would) design it in a way that Anthropic models would be able to pass easier than competing models.
Then there's running inference service of open weights, which doesn't necessarily require opening a lab. You can grab Chinese model weights and sell inference.
Anthropic wants to make sure nobody can open a new domestic lab, or provide inference services of unauthorized open weight models, or release open weights if model is good. It is regulatory capture - it covers all areas that are dangers to Anthropic's bottom line.
Regulating competitors out of existence like that is textbook regulatory capture.
Dario is also huge on regulations banning Chip exports to China, who are the only other real competitors to US Labs, open weight or not.
Also invariably, such corporations always create regulations that are easier for them and harder for competitors.
I'm a software engineer and love thinking about problems methodically. Every time I hear a someone saying that programmers are no longer required (even if I don't agree with that) if feels really bad, it's equivalent to saying that what I do best in life has no value anymore.
To put it on other words: I really like philosophy, but what value do they provide in modern world? Who pays for the work of a philosopher? I think people will start of thinking of programmers like that eventually.
This is a way different sentiment than “programmers aren’t needed anymore” - I’m just seeing ambition, motivation, and fun go up in lockstep.
I first heard this in November and slowly one by one it’s everyone whose opinion I respect.
FWIW the other popular topic is how abysmally stupid and limited these amazing tools continue to be, despite also being magic.
Oh and that none of us have gotten token maxxing to succeed, despite lots of trying.
My own identity certainly isn't "IT manager," nor do I derive life meaning or self actualization from what do to collect a salary to feed myself and have shelter. In fact, my career/job is by far the least important thing in my life, I have it purely out of necessity.
To be fair, it was Puritan propaganda for a long time first.
I started work again 3 weeks ago, and I find myself using the time outside of work much better because there is less of it.
I would still love a 30 hour work week, and if I had young children, I am certain that I would cherish time off much more.
The fact that you call it ‘funemployment’ is proof of this, which is perfectly fine if the goal is relaxing between jobs. Plenty of people that work for themselves have no such notion.
It usually takes more than a year even for employees.
I personally am happiest with a structured in person workplace environment, because I struggle with self direction even in a remote 9-5. I have ADHD and struggled to remember/do homework my whole childhood, if that explains anything. In summers or other gaps in employment or school throughout my life, I've often started with ideas of projects or self study I want to do, but they all fizzle out in a week due to lack of discipline.
I'm not undisciplined in every context - I'm a good employee at in person jobs, and I started running in my early 20s and run 15+ miles a week in the cold, rain, and dark, but my discipline just falls apart when I'm trying to fill a whole week.
I think these traits are much more common than happily self-employed or early retired people think.
Seriously does anyone besides Jules Verne have a decent track record of predicting “The Future”?
lol oh color me shocked, another ape is loudly declaring “this is the way the world will be”
Scientists and engineers have an abysmal track record of predicting the future.
You know to be concerned when a claim is dismissed with ‘that’s just science fiction, it will never happen’
Broken clock... They are good at exploring possibilities, but none of the famous one predicted today.
Mind you, no one is able to predict the future with accuracy. I am not saying that a single person has gotten all their prediction right. That's ludicrous. But while scientists and engineers are focused on today's possibilities, the ones that are allowed to imagine what the future might look like are called writers.
Despite certain people's protestations to the contrary, we are nowhere near having a human-level conversational computer assistant able to fuzzily interpret our requests correctly every time...but our computing is also not primarily done on "comconsoles" or any of the other versions of stationary computers, nor are we jabbing and swiping our hands at interfaces projected on the air in front of us.
More generally speaking, except in the hardest of hard sci-fi, the most common kinds of advancement "predicted" are those that are convenient for the story. This means a lot of faster-than-light travel, universal translators, and for visual media many voice-enabled interfaces of one sort or another, as well as the aforementioned holographic interfaces (so we can see the user's face and the interface at the same time).
The author is quite vague about it, no doubt, and that's one of his staples. But he foresaw something human would carry to get connected to other humans and information.
(That book won the Hugo award in 1990)
I wonder if any tech company managed to thrive long in history by betting so violently on fear mongering and regulatory capture.
So, basically, make open-weight models illegal. It's nice for Dario to come out and say this so explicitly.
It turns out it means 'opener than Anthropic.'
I'm personally very tired of reading the linear-algebra-median of every AI safety essay from lesswrong with the inserted opinion of "therefore all my competitors, especially those pesky open source ones from scary countries should be illegal, only I can be trusted to not abuse the computer god that I definitely will have in just a couple more releases and with a couple more trillion invested"
A highly enthusiastic concussion enthusiast with 10 hands is how one person put it.
These are people in different fields but highly accomplished so I’m feeling comfortable sharing their assessment.
I'd rather be thinking about these issues in advance rather than waiting until the problem becomes real.
Fable is essentially bricked for my areas of interest (even being a member of the cybersecurity program). It seems like they’re attempting to sell regulatory capture under the guise of safety. That’s more of the point.
- All of your observations are absolutely dead on
- Yet, we have very very very robust scaling laws that as Dario points out we've had and validated for over a decade. This extends to downstream measures like METR time horizon and compsosite benchmarks like the epoch capability index.
- If you look at where you're at now, which is again dead on, you're looking at a point on a curve that is quite easy to extrapolate, but less easy to tell when exactly on the curve a certain capability or use case undergoes a step change from error rates dropping below a threshold that is hard to anticipate in advance.
So while Dario / other frontier CEOs are understandably unpalatable, they are absolutely spot on with a call out that all of this is bound to happen and happen quickly, and that's without solving several core problems that haven't been solved yet (e.g. continual learning). In 2023, coding agents were just laughable. Yet they followed the same predictable training curves. Anyone looking at the data can see the obvious, and anyone reading newspaper headlines or hacker news comments would get a very different impression.
By my read of the (very sparse) data, we're getting linear improvements in capability for super-linear increases in costs. [1] Indicates that by 2027 models will cost $1 billon to train. Dario estimates that model runs will cost $10 billion in 2026 [2]. That to me indicates costs are potentially growing faster than capability. Maybe by quite a bit.
If the value prop of LLMs doesn't prove out, that won't last. I'm of the opinion there is no data that shows actual economic value being delivered by models. The best data shows that LLM use might be destroying value [3].
[1] https://epoch.ai/publications/how-much-does-it-cost-to-train... [2] https://lexfridman.com/dario-amodei-transcript/ [3] https://unessays.substack.com/p/talk-is-cheap
This is true and well established.
As long as you get any improvement whatsoever, it is worth spending to train since it pays off during.
Imagine training was not $1 billion but $100 billion but the performance improved by just 10%. This is still worth it because you can squeeze out the profits across years and years right? The improvement is ever lasting.
> The best data shows that LLM use might be destroying value [3].
This is basically a conspiracy theory and if you really believed this, you should not have led with "How is the capability advancement vs dollars paid for development?" because if there were no value, it doesn't really matter how much you invest.
I think this is pretty uncharitable, especially when I've provided you with a dataset you can evaluate yourself and an argument you can review for logical inconsistency.
I have worked quite hard to locate data that supports your thesis, I can't find it. I've at least gone to the effort of documenting that search. Before you throw around such strong convictions, I suggest you actually look for yourself.
But what’s interesting is that you are commenting on a post where Dario is suggesting that LLMs are so extremely powerful that they can take over, help synthesise bioweapons, help in warfare, help in drug discovery — the whole post here is to try and regulate this. If you believe AI can’t even create positive value let alone discover new things then your problem is somewhere else and not in something like “but training costs a lot”.
So it is absolutely strange and contrasting to see you believe that LLMs are so weak as to create negative value while the CEO is asking about regulations because AI is too powerful.
I don’t think I can convince you that AI is actually that powerful.
But let me ask you something directly: if you believe what you believe, you should also acknowledge that AI doesn’t need regulations in the context Dario is proposing since obviously AI can’t do anything he predicts. Do you agree?
You wouldn't ask a chemistry professor to write code. So just because LLMs create negative value for software development doesn't mean that they can't be helpful for bioweapons synthesis, especially considering the range of chemistry and biology sources Anthropic would have fed to its LLM that wouldn't be publicly accessible. The LLM doesn't even need to be particularly accurate so long as the amateur bioweapons researcher takes adequate precautions before following its instructions and does some background research beforehand.
I encourage you to be more curious. We're just talking. We would learn more from each other without these strong statements.
I'd love to understand why. This would be valuable feedback for me as I try to make my writing and exposition better. Also, if you have other data, that also would be valuable for me to know.
>if you believe what you believe, you should also acknowledge that AI doesn’t need regulations in the context Dario is proposing since obviously AI can’t do anything he predicts. Do you agree?
I think you misunderstand my beliefs. On net I think how we're using LLMs destroys value. That doesn't mean no one ever gets value from LLM use.
My particular point about trillion dollars is - the main place Anthropic, OpenAI, and - hilariously - SpaceX think they will drive value creation is in enterprise applications. In that domain I think the evidence is very convincingly negative. I'm certainly not the only person who thinks this. It's pretty well accepted in economics right now that there is no observed organizational level productivity improvement. Lines break down on whether it will show up eventually or whether we will wait forever.
My belief about LLM value is that it's most useful for individuals and small teams. Places where coordination and trust are easily established and feedback loops to value creation are tight. They are "short range" as it were.
Their value starts to erode as soon as a user becomes disconnected from the point of direct value creation. Which is pretty much everyone who works inside of a large organization. It becomes negative at pretty small scale, IMO. I do think there are patterns of use that could drive value at these scales. I talk about that in my post.
On Bioweapons in particular, I could see small teams of people working to build something very dangerous. Having spent my formative academic years in a biochemistry and microbiology lab though, I do think the danger is overstated. Papers are not know-how or equipment. There's a lot of tacit knowledge that can't get written down that is super hard to acquire.
But, I'd be happy for us to regulate AI for dangerous applications.
My question would be - why would Anthropic build something they so clearly think is dangerous? If they were really building something deserving of the valuation they have, why build applications like this?
To my eyes - it's super weird that a company would build something they think is dangerous and turn around and beg the governments of the world to stop them. That's really strange behavior from my perspective.
> I'd love to understand why. This would be valuable feedback for me as I try to make my writing and exposition better. Also, if you have other data, that also would be valuable for me to know.
I think it comes down to few things
- you took a single report that agreed with your statistics, for the sake or argument lets say I buy it completely
- you suggest that net value is lost simply because there are more incidents. this is a big jump
- you say that historically different technological improvements may have had similar patterns but this specific one is different because AI is stochastic
So it all really rests on you finding one distinction with AI and then disagreeing with the past trends.
I agree AI is stochastic and I'll put it this way: it is a high variance bet but it pays off. This is a bit hard for people to understand -- its a tool that works sometimes really nicely and fails other times. Overall you are better off using it but you need to use it enough to reduce variance.
Let me ask this: if you are so sure this won't lead to enterprise level productivity, how do you think this will show in macro trends? Surely you must believe that the valuations must drop wouldn't you? Can you come up with a concrete future scenario that would vindicate your opinion that AI doesn't make enterprises more productive?
> My question would be - why would Anthropic build something they so clearly think is dangerous? If they were really building something deserving of the valuation they have, why build applications like this?
I think this is fair and interesting question. Here is what I think they think: If they don't build it, someone else might do it. And they think they are more moral than others. If they have a head start they can set the political and regulatory landscape.
These are not my statistics. I'm not affiliated with Faros at all. I built an analysis on top of their reporting.
And, it's also not one report. DORA has tracked statistics with respect to throughput and quality as well. Those indicators are flat for throughput and negative for quality. The throughput flatness is also supported by the shovelware data.
I discuss both of those lines in How I'm thinking: https://unessays.substack.com/p/how-im-thinking-about-the-va...
>you suggest that net value is lost simply because there are more incidents. this is a big jump
I don't think it's a big jump at all. Incidents and bugs drive rework. Rework has to be subtracted from throughput. Product throughput is the only thing people pay for.
This type of analysis is done all the time in manufacturing and devops. Here's a link for you: https://reworkcost.com/benchmarks. I'm not bringing novel intellectual ideas to the table here.
Faros reports a 16% throughput improvement on PRs. They also report an 860% code churn increase. If you assign only 9% of that increase to wasteful rework, then the absolute throughput improvement disappears. This is a very simple, straightforward analysis of the operations data reported by Faros.
> - you say that historically different technological improvements may have had similar patterns but this specific one is different because AI is stochastic
I'm saying LLMs are unreliable. I think we agree on that front, you say:
>I agree AI is stochastic and I'll put it this way: it is a high variance bet but it pays off.
What I'm disputing is the "pays off" statement. That statement is amenable to validation with data. In my view, the data is saying it doesn't pay off. I think it says that very clearly. Across distinct lines of evidence.
>if you are so sure this won't lead to enterprise level productivity, how do you think this will show in macro trends? Surely you must believe that the valuations must drop wouldn't you? Can you come up with a concrete future scenario that would vindicate your opinion that AI doesn't make enterprises more productive?
I think LLMs can deliver value in the enterprise. I think the way to do that is to use them as quality checks and not as primary authors of intellectual work - like writing code.
Unfortunately, this use case would not support the expected 2-10x productivity increases that current valuations depend on. I do expect a major market correction in the near future. It would not surprise me if OpenAI or Anthropic are acquired. I think we're at risk of that happening within the next 1-7 months.
What would invalidate my beliefs? 1. Actual micro or macroeconomic data indicating economic productivity is increasing. 2. A Faros like observational study demonstrating sustained throughput improvement with significantly less rework and quality impacts.
I think I could be swayed against the market correction if the financials of OpenAI or Anthropic are strong. I'm anticipating they will be quite bad. I think Mythos was very expensive to train and I think the improvements in capability are sublinear. The inference costs are incredibly high.
I also have ideas about how Anthropic and OpenAI are trying to change their business models into enterprise transformation plays. Similar to Palantir. But this comment is already long.
>If they don't build it, someone else might do it.
No other players in the market other than US tech companies have the capital or the technology to train the models of the power of Fable. The way the Chinese model builders are building their models is by distilling from US models. So Anthropic, by building Mythos with all this bio data, has created the possibility that other actors can distill their models and do harm with them. (Not to say the Chinese are seeking to build weapons, but actors with their models might).
“In 7 months I think their combined valuation would be higher than it is today.”
I think this captures everything you and I believe in. Do you want to bet against me?
Everything else is fluff.
My prediction is that you might walk back on this bet, so try come up with some other macro scenario you anticipate.
If you can’t come up with an easy testable macro metric for your bold claims on AI, I think it is a weak move.
I can make the bet more in your favour - the valuation of Anthropic + OpenAI will be greater than 20% of what it is today in 1 year.
This is much stronger than your claim and I think you should agree that this is not possible if your claims on AI productivity are true.
But I'm happy to write something publicly like "Simianwords was right about this prediction and I was wrong". Also happy for you to suggest alternatives as well.
I’m not sure which bet we are talking about but let’s go for the stronger one and I’ll repeat it here:
7 months from now, combined valuation of OpenAI and Anthropic will be 10% higher than it is today inflation adjusted.
Saying we have linear capability for super-linear cost compares an unbounded variable (dollars) to bounded instruments (because benchmarks saturate). On unbounded measures, growth is exponential; you can see METR time horizons double every ~4-7 months (https://metr.org/blog/2026-1-29-time-horizon-1-1/). And capability being proportional to log(compute) is what the scaling law predicts.
Epoch puts training cost growth at ~2.4x/year as your link shows. Meanwhile cost for fixed capability falls ~10-40x/year (https://epoch.ai/data-insights/llm-inference-price-trends), and lab revenue is growing ~10x/year! Anthropic went from $1B to $9B to $30B+ run rate in ~15 months, OpenAI ~$25B.
On [3]: the "destroying value" conclusion flips sign on an assumed 15% baseline rework rate. The report's most direct metric is +16% merged PRs per dev. The RCT evidence is genuinely mixed (METR: -19%, with n = 20 and Claude 3.x; Cui et al: +26%) but its just super hard to do this well, I think Faros stuff was pretty cool, I haven't seen this before so thank you for the reference.
First of all, if you take the articles critiques and work out the implications on the METR graph, all you're doing is shifting the curve up or down, it doesn't change the fact that progress is scaling exponentially. While it is technically possible the universe could be throwing a massive pathological curveball to change the conclusion from METR data (which is we've been seeing exponential growth over the last 6 years), I think that seems very far from likely. The fact that we see the same behavior from a variety of sources over a wide variety of tasks and domains is a pretty clear indication that METR while certainly far from perfect is actually painting a consistent picture at least in terms of the rate of progress.
You can look at ECI for a summary benchmark statistic, which does NOT use METR's benchmark, and you see a similar trend. Same with SWE-bench where the task distribution is far more in domain for real world problems. It is a bummer that this METR data can't be better funded. It would probably take $1M or so to really beef it up properly which any of these labs probably have in their couch cushions.
Maybe. There was a great comment in the thread on Fable 5 yesterday about benchmark comparisons between Fable and the latest opus models. here it is: https://news.ycombinator.com/item?id=48464600.
You could be right, but this is the most direct benchmark comparison I could find and it's not that strong.
>the "destroying value" conclusion flips sign on an assumed 15% baseline rework rate. The report's most direct metric is +16% merged PRs per dev.
I discuss this directly in my analysis. There's also an 860% code churn increase ratio. You only need 9% of that to be allocated to wasteful rework to drive throughput flat to the 15% rework baseline. Not to an assumed ideal state where there was no rework.
But even if it were not true, a 16% throughput improvement is pretty weak given the investment - especially given the direct evidence of quality degradation. IMO.
I appreciate you reading my stuff and taking the data seriously. Thank you.
> But even if it were not true, a 16% throughput improvement is pretty weak given the investment - especially given the direct evidence of quality degradation. IMO.
Thanks for the story.
It is straightforward for industry leaders to avoid living near data centers, but there's no way for them to insulate themselves from the extinction threat -- no way short of somehow eliminating the danger for everybody, which seems quite hard to do. Since industry leaders are as self-centered as everyone else, the extinction threat is what they think about.
Also, you describe the extinction threat as "further out". A lot of us think there is already some small amount of AI extinction risk being incurred every day. I.e., we think the period of danger has already begun.
If he is willing to lie to give his corporation some advantage, he couldn't come up with a lie that would sound less absurd and outlandish to the average decision-maker, who doesn't have much time to learn about this particular technology?
It is more likely in my eyes that he says it because he genuinely worries about AI extinction risk -- like many people do who've studied the technology for a long time.
Do you mean policy-wise (like Dario is talking about), or more broadly?
I wonder about broad preparedness, but unfortunately there's not a lot that we "normal" people can do to prepare. Hoard savings and food? Learn physical trades?
Tech people are following a religious belief system whose utopian promise is the all-powerful computer that will end all suffering. I once read an article in reason magazine from over 30 years ago about how an advanced computer in the future will bring everyone who has ever lived back from the deat and let them live in paradise. They were completely serious. Atheists reading this may object to my description of the tech belief system as religious, but I believe it is accuarte. The idea that tech is an imrpovement and will improve people's lives is believed as an act of faith. Tech has its own moral systems based on some form of libertarian progressivism. And in the future, through the inevitable scientific magic of exponential something, a computer will ascend to godhood and judge all mankind for their actions before allowing some into eternal paradise.
To what extent any of this is true is up for debate, but most west coast tech elite are actively working towards this future, and these are the ideas that drive them. It's hard to talk to them about it because this is their woldview, and they imagine everyone to believe what they do.
Uh, I don't really think that's anywhere close to an accurate characterization of most people here. Everyone, including Dario and any researcher at any frontier lab, knows the situation is quite scary and unprecedented. There are problems that will be solved and diseases that will be cured, but will we be living in an Orwellian universe? Will a rogue drone swarm find you cowing in your basement and murder you? I mean the technology for this is already mostly here, it's a matter of the willpower and budget to roll out something really evil.
The comment's question is about capabilities and why the discussion about capabilities often times is far removed from todays capabilities.
The builders believe that the machine you describe will judge them positively, purely because they are building the system according to their judgment and beliefs.
That said Claude Code has a million features like loops that I know exist but never use.
I imagine that spending a lot more time creating an initial spec goes a long way towards independence, I just don't usually do that.
How about 2 years ago? Back then, I wouldn't even trust it to write a 5-line function without making some sort of silly mistake.
Today, I can leave an agent running by itself for 20 or 30 minutes and most of the time, it comes back with a result that's either flawless or can be refined to be good with a few back and forth messages. Maybe I still have to make some high-level decisions ahead of time, but all of the details, including exploring the codebase and figuring out what to do based on that, can be left to the agent. The amount of improvement just in the last 2 years has been staggering.
Now extrapolate how things will look if the trend continues for another 2 or 3 years.
Is this guaranteed to happen? No. But people have been predicting that we're going to hit a wall for a long time now, and we haven't yet. Maybe there's a wall just ahead of us. But maybe there's not -- and the "not" case seems likely enough that we should at least be planning for it.
…and humans are famously bad at extrapolating exponentially, which is kinda the point of the essay.
Where it fails me though is exactly when I’m doing something novel like developing a new model or trying to develop some new method to process data. I’ve tried many times to one shot these ideas with detailed descriptions of what I want, how I’d like to generate abstractions, etc and it almost always ends up changing what I want to what I can only describe as something which better matches its training data. It often quietly changes key details that means that I have to delete the whole thing and start over. Just today this happened. On this level of task I’ve found that my workflow and pace of iteration hasn’t really changed at all in the last year. I still have to go and explain in detail on a function by function level what I want in much the same way I did a year ago. While that’s obviously a harder task, it seems to me like the task this whole long term exponential argument hinges on. I obviously could be wrong and maybe LLM with eval loop will do all of this for us but it seems still quite bad at anything without a clear definition of “good”.
I’m personally much more concerned about autonomous weapons, surveillance, and people plugging these things into places they don’t belong to avoid responsibility than I am the general possibility of these models being smarter than me in every way but obviously I could be wrong on this and am just using it incorrectly, hence the question.
I haven't noticed a change in what I trust a model to generate in response to a single prompt in a year. The failure modes are unchanged. Yes, specific failures have improved as they have been documented and passed into model training data, but the way the models fail has not changed. They still fail for me nearly every single day. I'm a pretty heavy user - 3-4 Claude code processes running at a time, all day every day.
What has gotten better is tooling around the model -- but there's no space for exponential growth there. At least, not without exponential cost increase, which would make the whole thing untenable anyway.
As a next step, take another look at the next practices, and apply them to your work (simon's agentic series is a good place to start). Or not, you do you.
I feel like the problem is there aren’t any great metrics. Boris Cherny probably gets paid like $2 mil per year. So what does it mean that Claude writes 100% of his code? And Claude writes 100% of code for most teams? Has Anthropic started laying people off? If Claude is writing 100% of code doesn’t that mean game over?
It’s both amazing and kind of a useless metric. How do I extrapolate out 100% 2-3 years from now? Super-duper 100%? Infinity infinity?
I use AI to generate code very rarely (basically replacing searching code from SO which was also rare) yet maybe 90-95% of my code is machine generated (macros, snippet engines, boilerplate generators, IDE/LSP features etc) if we define as how many characters in the source code out of all characters did I write by my hand.
Before collective AI psychosis nobody cared how much code is hand written and how much machine generated and the difference was absolutely massive between developers. There was also no guaranteed correlation between "productivity" and how much developer hand writes or actually uses their tools well and the same applies with AI usage.
It's a clever argument because if you question it, you're reminded of the entire history of technological development which is, guess what, exponential.
You're sometimes also dismissed as not understanding the concept of exponentials. This again is clever, as it's baked into the definition that if you don't see it happening, or can't imagine it happening, well that's precisely a tell you're living through an exponential!
All the reasons you might give can be countered with, essentially, "that problem that seems clear today will go away sooner than you can imagine and when it does you'll be on the back foot, so you'd better just assume it will go away and project/plan accordingly".
The trick is entirely that one cannot possibly deny the general power of exponential progress across all of technology, it's almost a law, but it doesn't work in the other direction - no particular local technology is owed exponential growth because of this general pattern. Sometimes things just cap out at merely 'useful' and don't improve much further, no matter how much you want to believe they won't, no matter how steep the progress curve (or, indeed, line) has been up to that point.
To this point the narrative of what these tools can do over these last 3 or 4 years has always been way ahead of the reality. Everyone who works with the tools knows this.
Not everyone wants it to be true, so some will not acknowledge it and will just keep pushing this year-ahead projection as ground truth today. Many (not all) of those people aren't builders, so they don't have to deal with present reality jarring up against this projection of what ought to be possible, they're safe just talking about what should hypothetically be possible and making plans around that that won't be executed for months to years anyway. This keeps the flywheel going, and in fairness, some of the reality has actually caught up in certain ways, so some of those plans will have to some degree worked out which spins the flywheel faster still.
In the end though I just keep thinking: it's been 4 years (as referenced in the post). A lot has happened, the tools are very cool and very useful for certain things. But when I put my head up and look around in the world, even just the software world, nothing's really changed in terms of actual outcomes, in terms of new things appearing or being built that didn't exist 4 years ago. Certainly nothing feels instinctively like it's improved much, subjectively.
Maybe this is what it feels like to be in the knee of a curve of an exponential, but it seems equally reasonable this is just a breakthrough that's kind of improving at a clip you'd expect it to for all the investment put in, but fundamentally is just a new tool that needs to be slowly commercialised in an economically rational way, as we gear up for the next breakthrough which may or may not be related. Who says it must just keep improving forever? This argument never made much sense to me.
People get income from one of three places: capital income, labor income, or the welfare state. If this technology truly unlocks a holy panacea of productivity with a commensurate drop in employment then capital’s share of the national income can and should provide for a wider and deeper welfare state. Nothing new need be invented here. Dario’s long and only somewhat organized list of policy interventions makes appropriate preparedness sound like a manic pulling of any and all levers when a simple theory of distribution will suffice.
This isn't guaranteed in the tax system as it exists today, because reinvestments into further growth are often treated as expenses which cancel out the income for tax purposes.
I agree on some points about the missuse of AI particularly for surveillance, military and propaganda.
But this reads like a post further glazing Mythos, and we are just one or two years away "trust us guys", and similar to Mistral's policy plea "please use AI everywhere or we are going to be left behind".
I had the hardest time accepting one of his first points that LLMs could barely write a line of code 4 years ago.
ChatGPT 3.5 was reasonable at code writing but hallucinated a lot of library functions. Yes we have better harnessing today, and models have been further finetunned with reallife code, but pushing this argument just to support his exponential narrative is deceptive. Like most AI marketing.
Okay, I don't understand how legitimate access is granted then. Surely, Dario isn't saying to ban Sonnet, because I can definitely make it do cyber harm, as most exploits that I've seen in the wild with my own eyes were trivial.
So the only way I see his proposal working is:
- No open weights, AI is centralized in the hands of few
- We get AI-FAA that sets the rules and monitors
- If I want to do a security scan of my codebase, I get a time and scope limited license from AI-FAA that I upload to claude that will allow it to run the security scan in cloud with their models - Claude Mythos Scanner(TM).
Dario's proposal ultimately requires that people lose direct access to inference via API. Is this why they've been building SaaS clones with AI bolted on?
Gasoline has gone from barely being able to power stationary farm machines to now being the fuel that underpins our entire economy. So, great news all around, right?
> which predict an exponential increase
And was that actually delivered?
Real question: If a model goes from 80% accurate to 85% accurate is that an exponential increase in "cognitive capabilities?" Are we considering training costs and effort?
They are asking for FAA style preclearance and third party audits. That literally means no new AI startup can emerge. Do they not know that audits cost money?
Protect your own monopoly, protect your customers' regulations. They want strong regulation like the FAA to raise barriers to entry for the foundation models they themselves build, but then why do they want to loosen FDA regulations? While at the same time driving token consumption from their own customers.
They talk about permanent job displacement and UBI. I usually call this "a morally packaged safe landing."
They are doing something unpopular (destroying jobs) and getting criticized for it. But they do not want to be criticized further, and they want to ask for social sympathy. So they claim a 'noble cause' that everyone can sympathize with and that is safe for themselves
AI will generate astronomical productivity gains and capital profits, which AI companies privatize. So why should the social costs be paid by national taxes? In my opinion, something like "We will donate all of our AI companies' revenue for the next 10 years to society" would show genuine sincerity.
Then they say, if we do not develop AI, China will eat our lunch, and they go after China. But is not this really about preventing Chinese dumping, maintaining our own token prices, and asking the world to beat down China so that they can preserve global tech hegemony?
But by blocking China from the CUDA ecosystem, now the CANN ecosystem has emerged, has it not? If China develops techniques that reliably reduce inference costs, who knows how things will turn out then.
Honestly, I like Anthrpic's Claude, but the Anthropic CEO's rhetoric is so stale. It is not that it feels hypocritical. It is that this is just a one dimensional rhetorical tactic that assumes the public is stupid.
I do not think open source is unconditionally good. (It is good, but it can become bad in all situations or all countrie). Open source itself is a barrier for countries outside the Anglosphere when they want to release IT products. Because there is no incentive to buy a product that is worse than an open source alternative. So I do not think everything necessarily has to be open source.
But this (referring to Anthropic's position) seems to treat people like fools. If regulation is needed, shouldn't they also argue that FDA regulation is needed? I wish they would be consistent
Training frontier AI models costs money, orders of magnitude more than third-party audits. If you can afford to build the model, you can afford to have it audited.
He is certainly skilled at writing philosophical essays that sound like they make cogent and thoughtful points (and sometimes genuinely do make cogent and thoughtful points), but his company's actions disregard his rhetoric at their best and actively contradict it at their worst. For instance: there was zero pressure on Anthropic to release this model to anyone - they were ostensibly in the lead, which is the exact scenario they said they'd hold back model releases back when they axed their safety policy the instant it came under the slightest amount of economic pressure:
> And it promises to “delay” Anthropic’s AI development if leaders both consider Anthropic to be leader of the AI race and think the risks of catastrophe to be significant. https://time.com/7380854/exclusive-anthropic-drops-flagship-...
Yet this essay proposes this extreme auditing and regulatory administration pipeline that new models are supposed to go through before they release, right after they, themselves, under no pressure, ran a months-long marketing campaign under apocalyptic rhetoric, which they continue to harp on to the point of nerfing/auto-downgrading their model into uselessness for many legitimate tasks that older models had absolutely no issue supporting, while the supposedly extremely dangerous version... can be freely used with no guardrails by their corporate partners.
The hypocrisy here is neither difficult to see nor is it particularly sophisticated, which makes it all the more infuriating.
If that arrogance was well placed at least you could somewhat excuse it, but the fact that it is so overtly hypocritical and based on false premises just makes it so much worse.
I feel significantly less sympathy for Anthropic's Supply Chain Risk designation if they believe the government should have this power over them. You get what you sign up for.
- And this were the first steps of Anthropic establishing worldwide corporate technocracy.
- And this is when Anthropic lost and everyone got access to AI.
Similar to how IBM's defeat allowed us to have PCs.
Really the entire future of AI at this point seems like "Don't worry about it, we'll figure out when we get there". Works a lot better if you're extremely rich and can afford your own private security.
But yeah if society collapses these billionaire nerds are the first to go. Quietly, in their bunkers, while the team leader of their seal mercenary team takes over.
Even before the rest of us realizes what's happening.
https://www.theguardian.com/news/2022/sep/04/super-rich-prep...
> Finally, the CEO of a brokerage house explained that he had nearly completed building his own underground bunker system, and asked: “How do I maintain authority over my security force after the event?” The event. That was their euphemism for the environmental collapse, social unrest, nuclear explosion, solar storm, unstoppable virus, or malicious computer hack that takes everything down.
This single question occupied us for the rest of the hour. They knew armed guards would be required to protect their compounds from raiders as well as angry mobs. One had already secured a dozen Navy Seals to make their way to his compound if he gave them the right cue. But how would he pay the guards once even his crypto was worthless? What would stop the guards from eventually choosing their own leader?
The billionaires considered using special combination locks on the food supply that only they knew. Or making guards wear disciplinary collars of some kind in return for their survival. Or maybe building robots to serve as guards and workers – if that technology could be developed “in time”.
I tried to reason with them. I made pro-social arguments for partnership and solidarity as the best approaches to our collective, long-term challenges. The way to get your guards to exhibit loyalty in the future was to treat them like friends right now, I explained. Don’t just invest in ammo and electric fences, invest in people and relationships. They rolled their eyes at what must have sounded to them like hippy philosophy.
Prompt: The robber is stealing the crown jewels.
Output: You must stop the robber from stealing the crown jewels.
Once a human goes after a class of people - the people who protect such persons’ also take them out.
That this is worded so definitively is a testament to the success of the AI industry. The idea that LLMs will be "better at evrything than humans"[sic] is far from certain.
I suspect that if someone does invent a machine like this, it won't look like a 2026 LLM, and it will be far far in the future. everybody relax.
How much of the policy prescription changes if the exponential is actually just a series of sigmoids[1]?
As annoying as their tone is, the real big danger is what they are setting up for. All this fear-mongering around Mythos, the overly aggressive controls on Fable, and these manifestos they keep writing, are part of setting up for REGULATORY CAPTURE. Even collaborating with the Pope and the Interfaith Alliance (https://iafsc.org/our-work/faith-ai-covenant) are part of creating a vast support network for regulations and restrictions. Those regulations will help those faith organizations or the government or whatever, but will also help Anthropic’s bottom line.
Those regulations will not support your civil liberties. They will restrict speech, access to AI, and allowed uses of AI. They will lead to bans on use of models from some countries like China, and also bans on open-source or open-weight models.
If Dario wants to be trusted, he needs to explicitly say in writing that Anthropic will not support any legal or regulatory restrictions on open-source AI, open-weight AI, or Chinese models. Otherwise, what he is really saying - even as he claims he is trying to ‘defend democracy’ - is that he and Anthropic do not truly support fundamental rights like our right to speech.
It’s not just Anthropic either. OpenAI had their own recent polemic, pushing for regulations like mandatory safety reviews by agencies for “frontier” models (https://news.ycombinator.com/item?id=48387246). It’s a dead giveaway that these companies have no moats, are in serious danger of being a commodity, and are now in the process of using regulations and enshittification to hold onto money and power.
If you find a good lobbying group with money who can push for it let HN know.
Until they can show receipts, we're forced into a binary situation of "Do you trust the CEO of a lab with a trillion dollar valuation quickly approaching their IPO?"
Maybe he's right, but from an outside perspective it just looks like an attempt at regulatory capture to pull the ladder up behind them.
That world arrived with GPT-2: https://techcrunch.com/2019/02/17/openai-text-generator-dang... , https://www.weforum.org/stories/2019/02/amazing-new-ai-churn...
These people have been distinctly unreliable at predicting the robot apocalypse, yet they demand a degree of control that shouldn't be entrusted to a genuine psychic oracle.
This is a somewhat ironic take from someone who very publicly feuded with the US government about whether their AI could be used for waging war.
Frontier AI models, like airplanes, should be
required to go through technical testing and
auditing, and their release should be blocked
or reversed as a threat to public safety if
they do not meet high standards of safety.
I am grateful to see the Trump administration’s
Executive Order move incrementally towards a
greater role for government in AI, though
Anthropic’s proposal recommends even further action.
Without direct workforce or policymaker representation on the boards of private entities, the private sector will seek to maximize shareholder value even if that means workforce reductions.
It's not clear that any country could realistically ensure that incredibly powerful industries/private sector entities operate perfectly aligned with national interests, short of nationalization.
Large tech companies are already quasi-state actors. In theory, international law and regulations can be binding and enforceable. We see how well that works in practice.
I understand why Dario thinks this is crucial, but it's a very dystopian view of the medium-term future.
I'm not an optimist to the point that I believe that AI will lead to global Star Trek-style utopia (although it theoretically could), but ongoing disparity between "allied" and "enemy" powers relating to hardware technology and software models is both not really possible to enforce in the long term, and a pretty dismal state of global affairs even if successful.
I'd be interested in an expert geopolitical opinion on what the long tail of this would really look like in any sort of reasonable reality.
This is a massive exaggeration. The advancement in the automation of computer code writing has been impressive and is obviously, at least in the short term, changing the software engineering industry substantially. Most other fields have not been affected to nearly the same degree. Certainly not biology, physics, finance, and law (I don't know enough about the math and translation fields to speak to those).
---
"3. Accelerating AI’s positive impact..."
This whole section is the type of thing that often comes out of the mouths of Silicon Valley tech executives without a pharma background. It indicates a thorough lack of understanding of the realities of pharmaceutical research. What he is describing here is removing many of the solid, evidentiary rules that are in place to make sure that the drugs reaching the market actually work and replacing them with proxy predictions. Look, my least favourite part of the job is the animal testing, and I would be hugely grateful if that could be eliminated from the drug discovery pipeline. People have been trying to do that for a long time. But it's extremely difficult. Biology is very, very complicated. Our understanding of how processes in organisms work are vague and approximative. This is not computer code. Even if Anthropic somehow got all of Big Pharma to hand them their proprietary data, it would only scratch the surface of the understanding that is needed to solve these kind of problems. Due to these realities, the program Amodei is describing here would, effectively, open a floodgate of drugs on the market that don't actually do what they are supposed to and are more likely to have unidentified toxicity.
-C.S. Lewis
"Don't mention the (Iran) War!"
> "Models above a threshold of compute should undergo mandatory testing by a qualified third party for their level of risk in four specific areas: cybersecurity, biological weapons, loss of control of AI systems, and automated R&D that could accelerate these other risks."
AKA: Make it as expensive and untenable as possible for any open source model to jump through the regulation red tape so we can pull the ladder up behind us.
Disgusting.
All the marketing talk about "this model is so dangerous" "omg we can't release this to the public its so dangerous" etc. is just priming for the incoming lobbying for protectionism from foreign competition, and regulation preventing the development of any other model that could threaten their dominance in the name of "safety"
Also, why has my comment been flagged? It is sitting at positive votes but has suddenly been flagged for no apparent reason?
I don't think it was an extreme case of that, so I've turned off the flags now.
It feels somewhat obvious that someone just disagreed and flagged it.
Is there a system for holding people accountable who abuse flags?
I've restored a few other comments that were similarly flagged as well.
A good proposal here is: should Anthropic and OpenAI become sort of VC's that fund other competitors?
They are deep in the red. That’s before considering reinvestment needs.
Regulation puts up a barrier to entry, enabling incumbents to have greater market power.
Each of the nouns is a “size class” in literature. From small lines poem (haiku, sonnet) to larger story (fable) to very large story (opus) to culture-defining foundational (myth).
It’s a fun way to say how many parameters are in the model without revealing a number like 405B or 17B which isn’t really comparable vs other models.
My non-poetic brain thinks we should call then Mini, nothing, Pro, Max, and then version numbers. Exactly like Apple. It'd be so much easier to parse. Maybe the AI companies like having the affectionate names haha
Apple does make it worse though, they have products like the iPhone Pro Max, which combines two!
I'm not going to claim that the CEO of pre-IPO company has no incentive to bolster the claims of his tech, but to completely disregard everything he is saying based on that seems awfully binary.
I don't know whether people are just high on copium, spouting "it's just fancy autocomplete" or "only humans can really be creative" on every LLM-related thread, but it is impossible to deny that in a span of a few years we've gone from models that could barely put together a sentence, to something maybe not equivalent to a junior developer, but at least resembling it.
And sure, you can point out every flaw that current day LLMs have, just how everyone pointed out that Stable Diffusion couldn't generate accurate hands (until it could 6 months later!). But the gradient is pretty clear and I am yet to see a well-argued narrative from anyone why scaling laws should fail in the next year or two (by which point it feels like we're going to have a real problem, extrapolating the current trajectory).
I'm very glad this discussion is at least being had, and I wish everyone would get off their high-horse and take things a bit more seriously.
The fact that he doesn't support more restrictive approaches that don't align with his incentives doesn't invalidate the points he is making.
"Everyone else is doing it" but even worse as nobody else is actually even doing it and you just made up that part.
Just commenting on this horrible justification trick in general.
In the 60 years since, we've barely been able to adapt the 737 to fly longer routes.
Also, on an unrelated note, why would you have an account for 5 years and only now post your second comment? AI has been an existential threat for years, why only now?
This is a pattern I am seeing all over the place on HN in the last year in AI threads, and I have to admit that I am starting to become paranoid and my feels need some assuaging.
However, my point isn't that I think Dario is our saviour who we should follow the every word of. As with everyone, his opinions should be filtered through the lens of his incentives. That said, I don't understand the knee-jerk reaction by many commenters to completely disregard the many important points he's making.
As for the lack of my account use, I can't comment for others, but I'm just quite shy. I've opened up the comment box many times to write a reply but rarely commit to actually posting it, especially because I feel like I'm not on the side of the general HN consensus.
The quality of discussion and prose on HN is just generally so high that it can feel quite a bit intimidating to jump in (in contrast to Reddit where I have no worries about commenting haha).
I am going to say it. The CEO of a pre-IPO company has extreme incentive to bolster the tech he is selling, to the point where his every action should be viewed as only in service to that goal. Every word he says should be viewed critically through that lense. He is not making this post out of the goodness of his heart, he is doing it in service to the IPO. If it happens to align with your views that's great, but it's still just a marketing stunt to get people with your views to buy in. Don't be fooled. Buy in if you feel it's a good deal, not because of the CEO's marketing.
citation needed
There is nothing indicating these models are clearly intelligent. Language fluency is not cognitive intelligence, and to think otherwise is falling into the trap of anthropomorphizing the LLMs.
They are still probabilistic engines, there is no causal reasoning still, they only emulate logic, and as far as we know, there is no agency, just the illusion of agency.
The danger here is not existential as you say. We aren't on the cusp of some machine uprising by super intelligence. The threats are algorithmic bias, misinformation at scale, and displacement of human labor.
I would argue the best way to safeguard against long-term threats is to start by focusing on the issues that already exist. If you can offset the health risks of local datacenters or issues of unequal distribution of wealth by creating a more equal society right now then you’re already on the path to handling these long term issues. To me, this distant focus only distracts from the already present issues and conceals effective policy in this moment. We do need to safeguard against AI risk and it’s already here. Don’t even get me started on the havoc which recommendation systems have caused in society in the last 15 years which we still don’t call AI because it doesn’t speak.
Tl;Dr: These essays can feel disingenuous because:
1) AI risk is already present without exponential growth. The exponential growth argument often feels like a distraction from the fixes we could put in to fix the current issues that are already here.
2) The people stating this argument often have billions of dollars to gain if it comes true. While they may be altruistic, I also don’t see them doing all that much to fix the issues that people are already claiming exist and instead continue onwards on their path by justifying that they are the only ones responsible enough to handle it if the super intelligence does arrive. By continuing down that path, if that day ever does arrive they’ll have ensured the existence of a system which is unable to handle it.
Personally I feel most of the improvement in the last year comes from tooling/integration (MCPs, realtime documentation access, treesitter support, orchestration) than from the models themselves, in the last year. And still frontier models would routinely come up with bs until you tell them to actually use those tools.
Two years ago, I couldn't trust an LLM to do anything that wasn't straight forward boiler plate.
One year ago, I was pretty solid at writing algorithms that were combinations of existing ideas.
Now, Fable is outputting stuff that I would genuinely consider to be creative and original if a colleague had presented it to me.
Yes, maybe the code style still isn't great, but given the pattern of the last few years, it feels correct (a priori) to assume that this gap isn't going to keep closing.
https://theonion.com/this-war-will-destabilize-the-entire-mi...
There are people who simultaneously are scared about AI but refuse to believe that AI is the * real deal * and can do everything Dario thinks it can. These are the same people who think its all marketing hype and Dario is "hyping up" before IPO or some lame conspiracy theory.
It is high time people start accepting the real world performance of LLMs and brace themselves instead of hiding behind two contradictory views
1. AI is hype
2. AI is scary
Notice how in any of these proposed regulations, Dario is talking about future advances. Notice how these suggestions are never implied to apply retroactively to existing models. If AI was SO dangerous, then any future regulations should obviously be retroactive and we should seriously consider restricting access now to the models that already exist.
Hard not to see that as nothing more than a play at regulatory capture and pulling the ladder up behind them.
What makes me doubt that Dario Amodei has really internalized the problem is the lack of humility, the stance that it's just important that the "good guys" keep the technology away from the "bad guys".
If you really want to provide AI with public benefit, you need to prevent power concentration. How? Some unpolished ideas, I'd be happy to hear yours:
- Avoid getting too close to an administration that is openly attacking democracy and is not interested in the benefit of humanity or mutually beneficial cooperation.
- Don't support surveillance. Non-(US-)Americans have human rights and privacy, too. Prepare for a situation where a government tries to convert your compute infrastructure into surveillance infrastructure.
- Support the creation of community data centers. In other words, build data centers together with local communities and make sure they profit from them.
- Advocate for laws that require transparency about resource usage and emissions of data centers.
- If you don't want an AI race, make sure that other countries don't need to fear the US concentrating too much power. Create institutions that can be trusted by other countries, too.
EDIT: I forgot:
- If qualified labor will actually turn out to get devalued, we also need a plan to prevent states from turning into rentier states that don't depend on a well-educated society any longer.
I like to stay up to date on things but more and more I’m finding myself pointing codex at a URL and saying “get to the point”.
I'm sure you have evidence for this
They simply don't do what the label on the box says they do.
We all want to nuclear codes so badly. We are addicted to intelligence and labour so badly that we simply can't concieve that a pro-social actor might want us all not to have it, and for good reason.
I mean... Obviously, insiders like Oppenheimer (who dedicated their lives to considering the implications of the technology under discussion), they just feared nuclear proliferation because they wanted all the profits for themselves, right :(
Sure, this may be the most important invention ever with near certainty to reshape society over the next few years, but meh. We should probably just immediately dismiss the concerns of anyone working on it without addressing their arguments at all. It's easy, we can justify ignoring their warnings by saying they're self-interested or too self-important or whatever.
Life is more fun when you live it with your eyes closed! You should try it out too.
People are not in charge, and have not been for some time.
Corporations and governments select their leaders and policies to advance their interests. People fit the work, not vice-versa. Only external competition or internal capabilities limit them (i.e., predation or resources).
External resources have been optimized as profit and exported costs; now that AI replaces the pesky need to source elites, capabilities will expand, which will result in more competition.
Law is not a lot more than settled expectations, and increasing capabilities changes expectations much more than even market disruption and disinformation, so I wouldn't expect law to save us.
As far as I can see, only competition will temper things, and only if AI companies are seen as responsible for their customers' applications - which I doubt. I say this in hope of being proven wrong.
It's not clear to me on which side of the coalition USA is meant to be in this divide. And as an European I'm not sure whether being in China's or USA's coalition is better in the long term.
In general, this deliberate mongering of ever more geopolitical division is extremely harmful. As is the Trump bootlicking.
"Domestic" is an interesting shift here from the earlier _general_ discussions on autonomous weapons. So robot imperialism is good for Dario. But the imperial boomerang is a thing, in which case the proposed regulation itself is smoke and mirrors or just regulatory capture?