Europe solves problems

📚 PDF ⏳ 6 min 📖 7

Too volatile

This looks to me to be a nice place to live. Pool, outhouse etc, perhaps the main downside is it’s the middle of nowhere which might also be the appeal. Still 50% is a fair haircut in three years for something that is not supposed to behave like that. 

Volatility is one thing and liquidity is another. Bitcoin you can sell at the price at the click of a button and provided you aren’t selling billions, the market will meet you. With property, all the stars must align, right person, right moment, right money. It’s just not that liquid and if you need the money, you can really get skewered.  

I’ll save you the trouble. It’s south of Terrigal, a bit of a faff to get to, 90 minutes from the CBD by car. 

You could say everyone just went nuts in covid and this was the outcome but I remain in the camp of rural outperforms city in the next decade. Most of the improvements in technology are disproportionately beneficial to rural areas versus cities. Looking at Tesla's Robotaxis in America, they will also do well in Sydney and Melbourne. But we are falling over Ubers already. The net benefit is basically zero in a city. 

That is not true everywhere else. It is going to change communities across Australia for the better if it is allowed to happen. 

Remote wins, it doesn’t look like it in Killcare Heights, most likely because cleaning the pool somewhere that remote, with that many trees, probably costs $1,000 a week.

Perhaps the speed of change now in society is flowing into asset prices too. Very rapid accelerations and equivalent declines in industries that just get wiped out overnight. This whole process is just beginning. 

Errrr, yeah

If you are a pro-subscriber to GPT you already have Astra. There are several remarkable things about it. Perhaps most notably that it scored 99% on ARC AGI 3. The ARC problems are novel problems that have not been seen before. ARC AGI 1 was designed in 2019. The machines did not solve the ARC1 problems until December 2024, five years later. 

ARC2 was designed in 2025 and lasted six months. ARC 3 launched in March 2026, now it is September and it is completely solved. Keep in mind that OpenAI had this model internally around June, so that new benchmark lasted 90 days or less. 

Jensen thinks AGI has arrived. I think so too but only because I have not really noticed the difference between GPT6 and GPT5.6. My tasks are simply not hard enough to warrant GPT6, even if I think they are difficult. 

There’s only one show in town at the moment. Remarkably, less than 1% of the world’s population is using Codex though, the principal access point for the new model. The mega data-centre usage is a tiny number of people globally, who knows what happens when everyone wakes up to it. 

Millennium Prize

As part of the model release, OpenAI revealed that another model of theirs (not released to the public) had solved a Millennium Prize problem (more technically they ‘shared a solution’). The Clay Mathematics Institute set the problems in the year 2000, seven of the hardest unsolved problems with the most meaningful impact on maths. One is solved, six are not. 

OpenAI claims to have a solution to the Navier-Stokes equations which map the flow of fluids. This would be a foundational achievement……

….but, behind the story is rather more. Two mathematicians have also been working on the problem using AI. Tristan Buckmaster and Levent Alpöge. They themselves were making progress with a novel idea (now at the heart of OpenAI’s solution). They asked OpenAI if the model had access to their prompts, the answer was no. However, they did not receive an answer on whether this closed model had been trained on their prompts. 

Allegedly OpenAI offered to give them the $1m prize if it were solved and they were co-operative. It’s not exactly clear where this stands, but that background offer doesn’t exactly smack of innocence. 

Surely the bottom line is AI did the work. The progress even the mathematicians were making was very much AI driven. I guess Buckmaster and Alpöge will claim their novel idea was the driving force and will we ever know? I don’t think so. 

It’s instructive though because you can use AI to bring ideas to life at speeds not previously known. Like solving 90 year old problems in months. I feel for the mathematicians who were probably close, but this kind of knowledge wipeout is coming for many of us, perhaps not as brutally as this. But it is coming.

“Our” Economic Future

The following day, as if to hammer the point home, Anthropic released a report on the impact of AI on employment. Called “Our Economic Future” they meant everyone, rather than it will belong to them, which it might. Its worst case scenario shows “knowledge work” getting hammered and the general impact on employment being negative. I don’t agree that unemployment rises, where it did ring true though was the falling relative returns to labour. We have discussed that at length in this column, not so much that it is coming but that it has been here for a while and will likely now accelerate.

In their extreme scenario capital captures 55% of the economy versus labour’s dominant 59% now. Note though, labour is still better off but it won’t feel like that because some capital owners are going to get super, super rich and most people will feel worse off.

Overall it's a nice prelude to their IPO in which Anthropic staff and shareholders will get super-rich and we will indeed feel relatively poorer. Mostly, because that’s what we will be.

Still, I don’t think the extreme scenario is at all extreme enough.  BUT, there is something you can do. It is still a generational opportunity to take advantage of. Keep in mind less than 1% of the world uses Codex. The reason? They won’t pay $200, their employer is paranoid about access and data and so on. Overwhelmingly people are trapped in inferior AI products that are approved by someone that has no clue what’s going on and is solely focussed on not getting fired for a data or IP leak. I don’t see that changing either. Accordingly, here is what is intended to be a helpful list, if you do feel poorer in five years’ time, you can come back and refer to it to find out which ones you did and didn’t do. 

1. Codex. Download it. Pay for the $200 sub (even if only for a month) and get the app to open a spreadsheet on your computer and go from there. Your confidence will grow. Maybe it can do this? It can. And this? It can. 

2. Try an agent. GrokBot is amazing (again you have to pay). Open a new Gmail account. Auto-forward emails that don’t matter to the other account (AI will help you). Watch Grok deal with it.

3. Try a local model. Huggingface.co (harder - choose a small model). Will take you a day

4. Connect a free API. Try here 

5. Rent a server. About €10/month (and rising). You need to do this because you need an always on connection. Install Claude or Codex on it. 

Of course everyone is talking about ‘adopting AI’ but nobody knows what it means. Do the things above. 

Read the Anthropic report. 

Take the time to learn. 

You will say “I really don’t have time” and I will say “you really won’t have a job”. 

Euro-Trash

This is the Yara International ammonia plant in the Netherlands. A huge business that makes fertiliser and other chemicals for industrial purposes. The gigantic white tube is their carbon capture system. The European Union will stop at nothing to cripple their once world leading industries. One of them is the Emissions Trading System. 

In the case of Yara they must pay for every tonne of CO2 their business produces and their new scheme of carbon capture essentially avoids those fees, which run into the tens of millions. The story is one of European triumph. 

• Yara captures the carbon at its new carbon capture facility built with subsidy from the Dutch government. It is enormous. 

• Twice a week they ship the captured carbon on specially built vessels to Norway. 

• In Norway a heavily subsidised project called “Longship” then collects the carbon and stores it 2.5km below the sea bed. Approximately 800,000 tonnes of carbon will be dealt with in this way.  

Overall then, a manufactured carbon pricing scheme that costs billions, has caused a subsidised industrial response that also costs billions, with a subsidised solution, also costing billions.

It is a legitimate engineering success and I have no doubt, was massively complicated. Two European nations working together, engineering teams doing the same. But they surely solved a problem that didn’t exist. Anybody who makes fertiliser will not set up in Europe, the power costs are too high. The CO2 scheme is too ridiculous. 

There is a limit, of course, to which engineering projects countries and continents choose. Europe will point to future generations benefitting from lower carbon emissions. The population is voting on that though. Demographics don’t lie and they simply say that belief in Europe’s future is at all time lows. Birth rates are approaching 1.3 and it is because you can’t really do anything in Europe. Regulation will choke you unless you are already huge and can somehow skewer a subsidy.

It’s a very European success. Very clever, skilled and talented people built something complicated. They solved a problem their government created. It was only valuable because it was subsidised and helped avoid taxes and it might be good for the environment in an unspecified, future way.  

American LLM solves a Millennium Prize problem. European engineers bury carbon in a neighbouring country to avoid tax.

Further information

Our August 2026 report to investors can be found here.