Roundup #17: Europe vs Russia, some AI stuff and...good music
Just a good old roundup with some tracks I'm listening to a lot at the end.
Europe
One of my purposes in life seems to be resharing
’s writing on economics and politics on WhatsApp. My friends make fun of me for it sometimes. I’ve said before that his Substack replaced my Economist subscription. But not too long ago I felt that Noah was a bit too deterministic about the new Cold War / how it could turn hot. It seemed to me that it was overly simplistic to say that ceding Ukraine to Russia would inevitably lead to the fall of Europe (I’m exaggerating his earlier posts a little bit). So I emailed him:Same in Europe, if much of Ukraine is ceded to Russia, the probability of Putin reneging and taking the rest of Ukraine is significant, but not 100% and it likely will happen only after some pause. The probability of him then proceeding to attack the next country is also some x% but definitely not a certainty.
During this chain of events, Germany and the rest of Europe can, at any point, build a military industrial complex that would dwarf Russia's ability to build. Russia has the GDP of Italy.
So, it made me wonder if what you are saying is presented too linearly as a chain of inevitable dominoes falling. In reality I don't think that's really the case.
By the way, do you have any recommendations on the 'European Noahpinion'? I would love to also read a similar publication that is more Eurocentric, deeper into Germany/France and the machinations of the EU, etc.
He didn’t reply, but he did now write a full post about Europe, and it’s great.
The first key message is that “America is not coming to save Europe this time.”
That is the clear message of two landmark speeches from the past week — one by U.S. Secretary of Defense Pete Hegseth, the other by the Vice President JD Vance. Hegseth, speaking at a summit in Brussels on February 12th, declared that Europe is no longer America’s primary security focus:
Secondly, exactly as I wrote as well in my email: “Europe can handle Russia by itself — if it wants to”
Europe has both far more people, and far more manufacturing capability than Russia. When push comes to shove, I choose to believe that Europe would wake-up and start taking care of its own defence. In fact, I even have some hope that Trump’s re-election might create the much needed wake-up call for European leaders and people.
With the US under Trump maybe quitting, or at least pausing their championship of democracy and freedom, Europe might be the world’s last hope for human rights. But to become powerful enough to stand-up for democracy Europe will need to grow its economy and spend more on the defence (those are compatible goals!)
There are some signs that Europe is starting to realize it needs to make changes: Mario Draghi has been very active lately in arguing that Europe has kneecapped its own innovation abilities through ham fisted regulation. One small but dumb example is the cookie consent part of GDPR, Europe’s data protection regulation. Every website is mandated to get explicit permission from all users to store cookies. This is done through super annoying popups, where ~50% of people just ‘accept all’ anyway. So, as it stands, this law introduces friction online, cost European businesses about EUR 16Bn per year in revenue, while not solving any meaningful problem. I appreciate Europe’s focus on privacy, but they should have iterated on regulations like this to ensure they do something that outweighs their cost.
There is much to do, and many political obstacles ahead, but let’s go Europe!
I wrote on LinkedIn:
Imagine a self-driving car that works well 95% of the time and then 5% of the time it randomly tries to kill you. Would you get in? That's exactly how current AI agents behave, and why they are doomed without some groundbreaking solution to accuracy (reasoning does not seem to be that solution so far).
This is the deal-breaking difference between humans and AI: If you give a task to a human, they start with an initial solution (maybe not great), but with enough time and coffee they get closer to solving the problem. That's just how we work.
LLMs? Complete opposite. The first attempt is done in 10 seconds and often impressive, but then it gets weird. More attempts actually makes them WORSE at solving the problem. They tend to drift away from the solution instead of towards it.
Reasoning helps, but not enough. There's no "safe zone" where you can trust it'll work reliably. This leads to a clear conclusion: Current LLM-based autonomous agents just don't work. The human can't just be "in the loop" - they need to BE the loop, with the LLM as just another tool in the toolkit.
Another thought experiment to look at this: For any LLM (not talking about AI in general), there's always going to be a task complex enough that the LLM will fail, but a human will eventually solve it given enough time. It's a direct result of this divergence pattern I keep seeing.
Development projects
I finished the Canvas interface for Magicdoor.ai and it was difficult. Besides that, I have more product ideas that I want to build, but I noticed that I had some ‘blank project’ fear. Hesitation to start with a new, blank project. So I decided to do something that is easy and rewarding: rebuilding my personal website in NextJs. With my newfound development skills, things I wouldn’t even have known how or where to start on, like pulling a random quote from my quotes page onto the home page are now easy.
It’s like spending tons of time learning a really hard Van Halen guitar solo and then learning something like the Stairway to Heaven solo that once would have been impossible but is now well within my comfort zone.
I feel like singing the praises of NextJs, but I’m conscious that it’s almost the only thing I know. But I made my initial personal website with plain HTML, and that means you can’t do anything dynamic. Forget about putting a random quote on the homepage. If I added a blog post I even had to manually add it to the articles aggregation page. At one point last September I spent time rebuilding this site with ExpressJS on the advice of a friend. But it seemed like I had to create a database for the articles, and it was very cumbersome to even just add pages. With NextJs you just create a folder, and it becomes a page instantly. Then you can just throw a markdown file into that folder and the page is now a blog post. Then you can easily create an index of those posts, and it can be sorted, filtered, searched, etc, which means you don’t need to hook up a database at all. Fun and delightful things like making things interactive and animated are easy with something like Motion (very tempting to overdo it when learning to use a new toy).
It’s also, of course, a matter of spending time. By now I have maybe 200 hours of development experience with NextJs vs maybe 20 with EJS, but I kind of hated every one of those 20 hours while I really love working with NextJs.
Music
A chilled instrumental groove:
A great Donovan Frankenreiter’ish roadtrip song:
A slow-bluesy collab by The Rolling Stones and Lady Gaga:
Super underrated song by the Foo Fighters:
Very spicy guitar work from Guthrie Govan. Although it’s a bit pointless to debate who is the ‘best’ guitarist in the world, in my opinion if you tried to be as objective as possible about it, the only possible answer is Guthrie. I’ll go see him live for the first time on 14 March 🎸