I’m just wondering how this translates to computer manufacturers like Apple. Could we have these kinds of chips built directly into computers within three years? With insanely fast, local on-demand performance comparable to today’s models?
When output is good enough, other considerations become more important. Most people on this planet cannot afford even an AI subscription, and cost of tokens is prohibitive to many low margin businesses. Privacy and personalization matter too, data sovereignty is a hot topic. Besides, we already see how focus has shifted to orchestration, which can be done on CPU and is cheap - software optimizations may compensate hardware deficiencies, so it’s not going to be frozen. I think the market for local hardware inference is bigger than for clouds, and it’s going to repeat Android vs iOS story.
I feel weird defending Taalas here, but this argument is quite strange: of course it is more expensive now. It is irrelevant - all innovations are expensive at early stage. The question is, what this technology will cost tomorrow? Can it do for consumers what NPUs could not, offering good UX and quality of inference for reasonable price?
This is the same justification that was used to ship the (now almost entirely defunct) NPUs on Apple and Android devices alike.
The A18 iPhone chip has 15b transistors for the GPU and CPU; the Taalas ASIC has 53b transistors dedicated to inference alone. If it's anything like NPUs, almost all vendors will bypass the baked-in silicon to use GPU acceleration past a certain point. It makes much more sense to ship a CUDA-style flexible GPGPU architecture.
Why are you thinking about phones specifically? Most heavy users are on laptops and workstations. On smartphones there might be a few more innovations necessary (low latency AI computing on the edge?)
Many laptops and workstations also fell for the NPU meme, which in retrospect was a mistake compared to reworking your GPU architecture. Those NPUs are all dark silicon now, just like these Taalas chips will be in 12-24 months.
Dedicated inference ASICs are a dead end. You can't reprogram them, you can't finetune them, and they won't keep any of their resale value. Outside cruise missiles it's hard to imagine where such a disposable technology would be desirable.
Most consumers do not care about reprogramming or fine-tuning and have no idea what NPU is. For many (including specifically those who still mourn dead AI companions, killed by 4o switch) the long term stability is much more important than benchmark performance of evergreen frontier model. If Taalas can produce a good hardwired model at scale at consumer market price point, a lot of people will just drop their AI subscriptions.
Is progress still exponential? Feels like its flattening to me, it is hard to quantify but if you could get Opus 4.2 to work at the speed of the Taalas demo and run locally I feel like I'd get an awful lot done.
Bake in a Genius Bar employee, trained on your model's hardware, whose entire reason for existence is to fix your computer when it breaks. If it takes an extra 50 cents of die space but saves Apple a dollar of support costs over the lifetime of the device, it's worth it.
Yeah, the space moves so quickly that I would not want to couple the hardware with a model that might be outdated in a month. There are some interesting talking points but a general purpose programmable asic makes more sense to me.
I found the captions on Figure 1 quite interesting.
> Average performance (%) across four agentic benchmarks improves consistently with increasing model Intelligence Index.
> Centralized and hybrid coordination generally yield superior scaling efficiency, suggesting that collaborative agentic structures amplify capability gains more effectively than individual scaling alone.
Then again, the deltas between SAS and best performing MAS approach are ~8%, so I can't help wonder if it's worth the extra cost, at least for the generation of models that was studied.
> This isn't just compliance theater; it's a straight‑up national economic security play.
The woes of LLM contrasts…
In all seriousness, the points made ring true not only for European companies and should make everyone consider the implications of the current situation, as dreary as they are.
To me it is simply good writing, the kind that is found in literature, but feels a bit out of place in discussions on the internet. What makes it stand out is that real English speaker on the internet are way more casual in their writing. I’ve also noticed that non native English speaker are sometimes mistaken for LLMs due to these less casual sentences structure.
Yup, exactly. As a non-native speaker, I phrase things a little differently. Also, I’m used to using em dashes (from academia), but now that's considered an AI tell. Shit's dumb.
> Russian and China are already getting rid of Microsoft.
I don't know what you mean by China "getting rid of Microsoft" in the context of cloud providers. I mean, Azure is already present in China's internet, and just like any cloud provider present in China it's presence is a partnership with local cloud providers.
Russia is getting rid of Microsoft not because it has a choice. They are subjected to sanctions due to their invasion of Ukraine, and that essentially cut their access to all tech services. By that measuring stick, Russia is also getting rid of Boeing and Airbus.
The most interesting part is that they do not rely on Western software solutions (Russia still needs hardware, China may reach full autonomy soon enough). If they could do it relatively quickly, EU can do it too. And EU now has exactly the same incentives.
I heard him talk about GPLv3 someday, and what he said was that it was a mistake to call it "GPLv3", as if it was the evolution of GPLv2, because for him it should have been a totally different licence.
Which I find fair: there are different kinds of copyleft (like MPL vs GPL), it makes sense to say that GPLv2 is a different concept than GPLv3. Whereas I don't know if anyone should use GPLv1 because GPLv2 sounds like it fixed GPLv1 without changing its spirit.
GPLv2 was clearly intended to let you change the software on your devices. In some countries, GPLv2 already prohibits tivoization.
However, big tech found an exploit: In some countries, GPLv2 allows tivoization. This was not intended by the authors of the GPLv2. There was another exploit involving patent licenses, and a reverse exploit about license termination that allowed some developers to extort some users. They fixed these and made it the GPLv3. It's a bugfix release, not anything new. You only don't like it if you relied on the bugs.
Well, that's not really mutually exclusive with what I said. Those who called it GPLv3 consider it's a bugfix, those who decided to stay on GPLv2 consider it's a new licence.
True but obv. Only lunatics would use a Russian cloud service. The interesting part is whether and what extent China is different. Also, why Europe should start treating us like Russians.
I usually operate under the opposite world view.
If you've been personally affected by something, I no longer really trust you to be fair and honest and logical.
I don't want to hear about setting speed limits from someone that lost their child in a car accident.
Was that not an imperative statement agreeing with your cathartic comment? A little weird there isn't an explicit "this is why", but asking questions with a poorly conjugated why along with bad punctuation isn't usually a native speakers habit.
They said in a follow-up comment that they intentionally wrote something ambiguous, so… I don’t know, I wouldn’t waste too many cycles on comments that are deliberately unclear.
> The interesting part is whether and what extent China is different
Much worse for the EU, both strategically and economically. You’ll be able to buy Chinese services and give them your data and money, but you won’t be able to operate in their market. Germany is feeling the pain there. [1] Strategically they’re a Russian ally and are actively supporting Russia’s war against Ukraine and further aims against the EU.
Something like Russia -> China -> US as worst to least worst partners.
The EU should invest in technical and military capabilities and divest from reliance on other countries and echos the US American position very closely.
Those can also be explained by favouring usa/venezuela oil while still supporting Russian politics. For example: in the Ukraine war he is constantly seeking ways and arguments to support putins position even though he is one of the few leaders worldwide who do this.
They're not favouring - they're even enforcing secondary sanctions on Russian oil on China and India, which is difficult and expensive.
The fact that you can divine pro-Putin things from his speech means nothing compared to him performing massively powerful economic action against Putin.
The Trump administrations political positions are effectively a one to one match with https://en.wikipedia.org/wiki/Foundations_of_Geopolitics which Putin aligns strongly with. Knowingly or not, Trump has been a tremendous asset to Russian geopolitics in every sphere.
You are ignoring Trump literally adopting Russian proposals and demands in that war as they are. Trump and america flipped sides, seeing Russia as admirable peer and Ukraine as someone who should shut up and put up.
Trump want deals with russia to enrich himself. For that he needs Ukraine to loose. Bad thing is Putin does not have enough, he wants the rest of Europe too.
Just as America would like to reduce its dependence on external production, so to do other countries want to reduce their own. We used to live in a world converging toward maximal international trade, when in fact it was exploiting underdeveloped nations. As we progress globally, and as the development gap shrinks, we have noticed power dynamics which weren't well guarded against in the old way.
So now what? How do we preserve a lot of the efficiencies of the past, while strengthening the resilience and redundancy. How can multiple nations create policy which drives business on partially compatible protocols?
If I allow myself to be optimistic, I'd be hoping for more international lawyers and trade agreements. Protectionism is natural, but taken too far, isolationism is a death sentence.
> we have noticed power dynamics which weren't well guarded against in the old way
The clearest example is a dependency on a single wealthy nation for military and world policing. It's a good thing for individual countries to be able to project their own foreign policy goals like containing Russia without having to rely on the whims of another country's politics. Even here in Canada we should be able to defend their own arctic border reliably and be able to project power to China/India beyond strongly worded letters.
> I'd be hoping for more international lawyers and trade agreements.
Ignoring the US's recent moves there does seem to be more trade deals than ever between 'middle powers'.
> isolationism is a death sentence
The best way to maintain global relationships is to offer tons of value. Similar to how China can get good trade deals and influence simply because they have so much to offer economically. This isn't just issues of diplomacy.
One of the USA's greatest exports is intelligence and higher education, and what has been happening with that and the general anti-intellectual atmosphere is to me the most concerning as an american. Ironically, public education in america has been pretty bad for a while. But I'm going to start rambling here... way too many problems, and no damn leadership.
A country were a lot of its citizens don't have access to basic human or social needs, and equate a demand for that, already available in the rest of developed world, to "far left political activism" - that is really ironic. There is nothing left on the left (pun intended) in today's America.
I don’t know if this counts as activism, but I was at my university’s Faculty Club and a faculty member walked over and immediately started bitching about Donald Trump without introducing themselves. Like, you’re supposed to be in the business of developing people. What a gigantic waste of time and money.
This just sounds like basic social bonding for red blooded Americans these days. Would you have condemned them for similar commiserating in the aftermath of September 11 or Oklahoma City?
Nearly every Republican who was in Congress in 2014 would have described at least 3/4 of what Trump has done this term as illegal and totally unacceptable, and would have described at least half of the rest as incompetent.
Unless you can make a case that since 2014 the country has moved so far to the right that even 2014 Republicans are now "far left", about the only thing you can infer from someone bitching about Trump is that they are probably not far right. (Even that one is pretty iffy because he's pissed off a lot of the far right now too).
If they were instead lauding Trump would you also see that as a waste of time and money? If they aren’t doing this in the classroom I don’t see the issue.
Sports teams and "after school activitie" are a much much higher priority than teaching. It isn't even close. It seems the only thing we prioritize in education is... entertainment? I'm sure that will be GREAT in a few generations?
> I think at the moment Canada faces way bigger problems from the south.
Exactly. Not only there's the absurd campaign from the Trump administration on how Canada should be a state but there are also the recent treasonous talks between representatives from the Trump administration and the Alberta separatists.
Putin's regime might be a cancer of humanity, but Canadians have far more reasons to feel threatened by the Trump admin than from Putin, even if Trump is a proxy for Putin.
> like containing Russia without having to rely on the whims of another country's politics
That's true, but at the same time it was probably already the case before invasion of Ukraine, and it is definitely the case now.
The main issue is political fragmentation: would Paris and Berlin risk lives of French and German people (soldiers and civilians due to retaliation) to save Vilnius?
But if the answer is true (as obligated by the Treaty of Maastricht, independently of NATO) then Russia stands no chance with conventional weapons against the whole Western Europe, the balance of military, demographic and industrial power is ridiculously lopsided (involving nuclear weapons would also raise the same political question about the French willingness to nuke Russia in retaliation to Russia nuking Poland but if the answer is yes, Russia cannot win a nuclear war either (which everyone would lose)).
> The main issue is political fragmentation: would Paris and Berlin risk lives of French and German people (soldiers and civilians due to retaliation) to save Vilnius?
This is a wrong question. If one day Russia feels brave enough to attack any NATO country, the right question to ask is, "Do we want to fight this war on someone else's soil or on ours?". This is the reason why Europe is so focused on helping Ukraine BTW.
What do you mean? There was never any question of attacking Russia and fighting any war on their soil. Nobody in their right mind would attack a country with the 2nd largest army and nuclear weapons. The war in Ukraine definitely made this army still very weak, but, except Ukraine defending itself, I don't see anyone rushing to attack Russia anytime soon. It makes no sense now and made no sense before they invaded Ukraine. There is nothing to win by attacking Russia and a lot to lose.
The best way to 'attack' Russia is to undermine its economic and political systems then let unrest amongst its citizenry do the dirty work. 1917 showed Russia's proletariat was very effective at achieving regime change.
How do you undermine the economic and political system of a country? The economic one can be undermined by sanctions, and they happened only because the war - before that the West was happy to send billions to Russia. The political one seems quite stable, Putin had a few decades to cement it and make sure nobody takes it to the streets, and if someone is brave enough to do it, they will be quickly pacified. He is switching the internet on and off and there is no sign of Russians reacting like Iranians.
But I saw several people criticizing their relatively high position on this chart given high incompetence and losses.
EDIT: Apparently this website doesn't follow any rigorous methodology. So basically the only thing their army is 2nd in the world is the nominal number of nukes (hopefully most of them don't work).
> This is the reason why Europe is so focused on helping Ukraine BTW.
We aren't that focused actually. France produced close to zero 155mm shells in 2023 because the producer (les forges de Tarbes) couldn't pay its suppliers due to liquidity issues. That could have been solved by a phone call to the national investment bank (BPI) but lasted 9 months because we don't take things too seriously.
Another example is how negligible was the war in Ukraine in the debate about the government budget for the past two years. If we were serious about helping Ukraine we should be spending so much money it would become a priority topic in budget debates, but it's not the case at all.
I'm deeply disappointed about how complacent we have been for the past 4 years.
The answer is always going to be "maybe", but hopefully enough of a maybe to deter hostile actions. That puts everything in an uncomfortable state of uncertainty.
I wouldn't be so categorical about France. Pro-russian/ “anti-war” political parties earned the majority of votes (but not seats) in the last elections, and the personality of Macron is so divisive (he has had record low approval for most of his tenure) it really impairs support for war.
Right. Aggression can only be tolerated up to a point before it triggers a response. Remember, on 1 September 1939 the Nazis invaded Poland and two days later both the UK and France declared war on Germany.
“and two days later both the UK and France decided not to intervene and just set up defensive position in Belgium and eastern France” is what actually happened. With the terrible results we known for France (the defensive position being hammered on its weakest point, leading to the complete collapse of the French army in less than a month.
> Just as America would like to reduce its dependence on external production, so to do other countries want to reduce their own.
If anything, I'd say for other countries it's more urgent.
If China embargoes deliveries of light bulbs to Europe, all the light bulbs already in place keep working. The pain would grow over time - giving a grace period, to ramp up local production.
If America embargoes AWS, Google, Apple and Microsoft? The pain would be instant and severe.
That would be as close to a declaration of war as you can get without firing a bullet.
The immediate and obvious response would be for the foreign branches of those companies to be declared "of national interest", nationalized and forced to keep operating.
I'd imagine the government would be in talks with the highest ranking local Amazon employees long before, but I can't imagine a country trusting the hardware or wanting to manage the jank.
AWS China is a completely separate partition under separate Chinese management, with no dependencies on us-east-1. It also greatly lags in feature deployments as a result.
I'm absolutely not an expert, but critical things for power and food production not to mention medical supplies and emergency equipment are also tied up pretty deeply in international trade.
The world would break pretty quickly if we all just stopped trading with each other.
Sure, but many products can be sourced from a load of countries.
If you can't get natural gas from Russia you can get it shipped from America or Australia or Qatar - it's expensive as hell, and you might need to quickly build new regasification plants, but your economy keeps running. And there's no remote kill switch that disables the gas you already have in-country.
That's not the case for the services provided by AWS, Google, Apple and Microsoft though - the 'competition' is one US provider vs another.
In case of war AWS, Google, Apple and Microsoft and others would be immediately directed by government to adopt its war strategy—like it or not—just as US manufacturing was forced to retool for war production during WWII.
It is risky to believe that the development gap alone makes for higher economic efficiency when manufacturing things in China. There are very real structural differences in how various industries are organized. Not least in terms of geography.
This is an aspect the west seems to have missed entirely as there are no attempts to learn from it or emulate it.
Everyone knows about Shenzhen. Not everyone knows that this is how every major manufacturing industry is clustered in China in various cities and regions.
My point was that the development gap is what lead to the current situation, not that it's just cheap labor that makes Chinese stuff cheap.
My point about maintaining higher economic efficiency is actually the same point you're making. How can the globe (not just the west vs the east) learn from the past and build for the future. We live in a magical world with translation services available to billions of people, how can we empower them to organize around the right ideas. How can we preserve culture and art while flooding ourselves with technologies developed globally? Who pays for security and research? Intellectual property law in general?
So many big issues and questions still need a lot of work.
I think Congress is actually the biggest obstacle to efficient manufacturing in the US. It is a body where the primary motivation of representatives centers around what they can get for their constituents, not what makes sense nationally. So any government spending (eg procurement) will actually tend to drive fragmentation as representatives fight for their states.
Take for instance the space sector. It is fragmented by design. By Congress. Not only is it spread all over the country, making collaboration expensive, time-consuming and clumsy: there are essentially six different federal space agencies of which NASA is just one. This is terribly inefficient.
I remember when reading about the Apollo missions it was astonishing just how much time they lost by different parts being built all over the US and then shipped across the country to be integrated. Utter engineering madness that was only made to work because one could pour immense amount of cash on it.
This is why companies like SpaceX was able to run more efficiently: they do a lot more in vastly fewer locations. Ditto for Lockheed during their golden years: Skunk Works was famous for having "everyone under one roof and within walking distance of each other". (That Skunk Works neither exists anymore, nor can it exist, but that's a longer story which is also about extreme inefficiency).
It is reasonable to assume that Europe wouldn't do any better. Or any setup where politicians are inclined to optimize for regional gain. You'd probably end up with the same political fights over who gets what if the EU were to push towards more of the kinds of manufacturing that you find in China.
We know we're inefficient and we have some idea of why. We like to blame factors that are easy to politicise or evoke emotion (environment, exploitation of the poor etc), but I don't think they are as important as people tend to think.
We just don't want to change. And there are legitimate reasons for that. Chief among them that we're uncomfortable with strong central control. (Well, we used to be. It only took a majority of republicans about a decade to turn 180 degrees on that question and prefer an all-controlling federal government dominated by the executive branch)
I think Susan Collins is a great example of this. Her support of Kristi Noem is based on deals she finds acceptable for Maine residents. The fact that other states suffer at the hands of ICE doesn't effect her decision's. Collins feels she only is responsible for Maine and not humans that live outside of Maine.
I find this sort of compartmentalization offensive to the common good.
That's representative democracy for you. Heck, even China faces the same issue, but they get to make it a competition between provinces, on who can win the favor of the emperor. Helps for them that the emperor has supreme authority though.
No, that's the incentive this specific system creates. There are democratic systems which do not suffer from such hyper localism. Such as the German mixed member proportional system.
Sounds like a narrow interpretation for representative democracy:) Maybe I'm stretching/mangling the golden rule but "do unto others as one would like others to do onto Mainers."
That's only in peacetime. During WWII the Government directed US industry to gear up for war production and the transformation was not only remarkably swift but also the largest retooling effort in history.
In these fraught times it's worth revisiting that history to remind ourselves of what's actually possible. By today's standards, the US's industrial response to war was truly remarkable.
The US did this with automobile and steel industries concentrated around the Great Lakes. It's not some kind of profound insight on the part of the Chinese.
The downside is that it decimates entire regions if/when the demand for what they produce drops.
Yes, it has its risk, but that isn’t why the US or Europe don’t cluster industry to create higher efficiency. The risk can be mitigated. The political willingness and ability to do it deliberately just isn’t there.
> How do we preserve a lot of the efficiencies of the past, while strengthening the resilience and redundancy.
Open source with clear international governance and maintainer/contributor base, in such a way that a geopolitical rift leaves both sides with working software.
That works for tech and the infrastructure, of course, but not for the corporations built upon them.
> more international lawyers
I don't see that as a significant source of safety in our current world.
> isolationism is a death sentence.
The current US admin isn't isolationist, it's merely reverting back to 19th century imperialism.
>Just as America would like to reduce its dependence on external production, so to do other countries want to reduce their own.
Conflating the president's desire and projecting it onto ordinary people. Most people don't care about this issue, it's the current president who is hellbent on destroying free trade.
> If I allow myself to be optimistic, I'd be hoping for more international lawyers and trade agreements.
One of the issues with the current system is that the WTO appellate body, which is effectively the court of world trade, requires USA approval for any appointments, which both Trump and Biden have refused to give. This effectively makes the WTO completely impotent.
Trade agreements, the WTO, its rules and appellant system, only work if nations are at peace and that peace is sustainable. We've just lived through a remarkably stable period of 80 years (since WWII) without which WHO, free trade and trade agreements could not have existed as we've known them. That era is seemingly now over, and the WTO is falling into irrelevancy.
Unfortunately, in the decades since the 1970s laissez faire economics/capitalism with its immediate need for quick profits, short-termism, a penchant for deregulation and ignoring traditional business ethics has meant that governments have ignored their long-term strategic interests. Despite the dangers of these policies being blatantly obvious dangers from the outset many Western governments encouraged such practices. Now it's payback time, and it'll be expensive—likely more than if the old order had been retained.
Anyone with a sense of history could see the headlong rush to deregulatate markets, indiscriminate reductions in tariffs and free (and indiscriminate) trade, would ultimately result in leaving many countries strategically vulnerable and open to exploitation by others.
We're now witnessing the true cost of these policies and what it means to have lost critical industrial infrastructure, loss of production know-how along with the loss of skilled workers, and an ongoing deskilling of the workforce all of which took decades if not centuries to build up.
With more nuanced policies much of the pain could have been avoided.
Rebuilding a strategic manufacturing infrastructure to insure resilience and independence in an increasingly uncertain and divided world will be costly and difficult.
Even without national protectionism we are still experiencing isolationism, expect instead of it being done by nations in the interest of their citizens it is being done by corporations in the interest of their shareholders and it's leading to a dangerous amount of centralization as well.
Compatibility protocols are probably the best answer, allow individual countries to develop software they trust to interact with internationally accepted protocols and formats. As you said, good luck getting anyone to agree to anything. If email didn't already exist I don't think it would even be possible to implement today.
>"but taken too far, isolationism is a death sentence"
I would argue that few large countries have everything to be self sufficient. For the rest - they would have to band together to avoid being at the mercy of their bigger overlords.
As for efficiencies of the past: I think they lead to a complete monopoly / near monopoly in few critical areas. The result - the monopoly power becoming a political weapon and or critical vulnerability.
It might seem contrary to the current trend, but I've recently returned to using nvim as my daily driver after years with VS Code. This shift wasn't due to resource limitations but rather the unnecessary strain from agentic features consuming high amounts of resources.
The frontrunner was an independent candidate who claimed to have spent nothing on his campaign, asserting that it was entirely run by “volunteers.”
Romanian secret services, under the directive of the current president, released reports concluding that both state and non-state actors had been involved in manipulating public opinion.
The candidate holds extreme right-wing views aligned with Romanian neo-Nazi groups. He has repeatedly referred to Romanian Nazi leaders as heroes and has expressed admiration for Putin, calling him a hero as well. His speeches often included mystical elements and rejected modern medical science, denying the existence of viruses and questioning the effectiveness of chemotherapy for cancer. While he has made other controversial statements, I will leave it at that.
The candidate seems to be more of a communist than anything[1], he wants 51% state control of all large corporations operating in Romania. Labelling him as right-wing is asinine. Actual communist parties are supporting him.
Perhaps if people wouldn't demonise any anti-globalist public figure as "far right" we wouldn't have ended with this clown of a candidate. People haven't voted for him as a person, he was totally unknown as of a couple of weeks ago, they've voted against the status quo, and more importantly they voted against the establishment that regulates what constitutes acceptable opinion.
In the page that you linked, it states that he wants to implement a measure that enforces minimum 51% participation by the state in all natural resource exploitation activities on Romanian territory.
While this can be viewed as a left-wing policy, it can also be a form of economic nationalism.
As far as I understand, he gets classified as far right [1], because of his ultranationalist and ultraconservative views.
The parties that declared support for him [2][3][4], after the first election round, have similar views.
Web and mobile developer, my focus is on helping you create your product.
My last long running project was a cross-platform app that was built using Angular, Ionic, Python, Flask, Celery, MariaDB, Elasticsearch, Redis, FreeBSD.
reply