Guys, they made the trolley meme with a boomer vs 10 young people. Actual genius meme-warfare move.

Update: @MosiniElisa nicely informed me that the initiative was to let MEPs individually select art that they wish to display at the Parliament’s building. This was selected by Danish MEPs… already much more defensible. But then again: instead of this, the Danes could play a pivotal role in making Europe a geopolitical player capable of deterring claims on Greenland. (And many European countries need to copy-paste much of the Danish governance systems)

People who say climate change isn’t happening are living in la la land. You can see it. A friend of mine showed me an old picture of his father in front of a glacier in the Alps. He went back to the same spot to recreate the photograph 40 years later. The glacier was gone.

Interesting take. This could really be a reason: building data centres is a relatively new expertise, combining both older and new fields and is a highly lucrative industry. Boomers must watch young guys out-earn them while doing useful work they don’t understand. Infuriating.

Good to know my tax money is spent on grotesque art trying to somehow “diss” the Orange Man instead of actually doing something about Europe’s problems.

¿Alguno de vosotros sigue atento a lo que pasa en Red Eléctrica? ¿Os acordáis del Gran Apagón y que nadie dimitió por ello? Desde Ave Europa hemos perdido la cuenta de las veces que hemos estado al borde de otro Gran Apagón. Y la causa está en que tenemos la red eléctrica nacional al límite. Ahora mismo, "enchufar" una fábrica nueva pone en peligro toda la estabilidad del suministro eléctrico. Añadamos eso a la poca estabilidad característica de fuentes eólicas y solares. Estabilidad imprescindible para el funcionamiento de todo el sistema eléctrico. Estabilidad proporcionada por la energía nuclear, la hidroeléctrica y el ciclo combinado, además de sistemas auxiliares. Desde Ave queremos una Red Eléctrica debidamente mantenida y escalada a nuestras necesidades, así como ampliar la nuclear para no depender del gas natural y aumentar las presas hidroeléctricas capaces de ser llenadas con los excedentes de las renovables. ¿Qué propuestas has visto en los actuales partidos sobre energía? En Ave proponemos soluciones, no demagogia. vozpopuli.com/economia/energ…

An ad I saw today on LinkedIn. I like it. No nonsense, clear message: “come to Finland, we’ll make it easy for you to build stuff”. I wish we all had this mentality.

I have been to Texas several times because of family, and Texans don’t give a damn what the solution is as long as it solves the problem in an economic efficient way. Also, lots of sun and land. Ergo: solar + batteries 📈

Very good article. I’m not an expert but I’ll give my take nonetheless lol. One of my “hypotheses” has been for a while that AGI, Skynet and Deus ex Machina are in fact *not* around the corner. No matter what Dario, Sam and Elon keep saying. I am fully aware that these individuals are much more intelligent, informed and hard working than me, however I also believe that they have gorged on capital by promising to create God and so they have strong incentives to always keep promising more crazy stuff “next year”. If AGI is not around the corner, I think the smart move is to focus on defensible and profitable “niches” (niche as opposed to general, we’re still talking billion dollars markets lol). I still think the one of the easiest and most profitable niches will be programming, just by looking at the job market for software engineers: it’s a highly demanded, expensive skill, so making software engineers 10% more productive is a big deal and may make the entire effort profitable. It’s the “easiest” because code lends itself very well to the LLM text-focused architecture. I don’t think there’s a significant chance of anyone catching up with the big American firms in this niche. Beyond that however, I think the defensible and profitable niches become *way* harder to predict, because other engineering disciplines are much less text-based. “The thesis is that models trained on company-specific data and physical-domain data can outperform American general-purpose models on engineering and manufacturing tasks”. I understand the thesis and I genuinely think it’s a risky bet. It’s not clear to me that LLMs will be able to reliably parse the embedded information in 3D modelling CAD files, PCB layout files, electronic schematics files etc… arguably, neither can a human, which is why you also need to document design decisions, test results and in general have a knowledge-transfer system in your company. An electrical engineer I work with told me he was still very disappointed by AI’s PCB design performances. An aerospace engineer I work with has stopped using AI for mechanical design. Mechanical engineering often requires solving an entire new problem in 3D with some form of novel mechanism (here evolutionary algorithms might be more suited than LLMs but they have other issues). The whole digital twin thing is far from new (I actually worked on this as a student lol), same for the logistics optimisation through digital modelling, both are data-intensive but not necessarily compute-intensive, and as standalone techs are not really dependent on AI. It will still be a profitable niche because large companies are very reluctant to move their data around and it will help cut some waste, but I think the real advantage will go to the first companies that really manage to integrate these data-heavy contexts with complex modern AI… what those use-cases will look like is way beyond my pay grade lol. The partnership between Mistral and EDF is an interesting case of this. Yes those companies generate lots of industrial data, but it’s effectively mostly “just numbers”… I imagine a large percentage of EDF’s data is effectively spreadsheets, for which the good ol’ “Data Science” (before it got also renamed to AI) was already good enough. In that category, in my small personal experience (my first full time job was in data science for industry), the bottleneck is usually neither the data nor the model, but the combination of the various domain knowledges (for a chemical plant: the chemist, the electrician, the civil engineer…) with whatever algorithm and data. Or if you prefer: the bottleneck is still the interaction between humans and computers. Anyways, this will be an interesting field to follow closely and lets me think I should not have difficulties finding a job in Europe in the next decades!

It seems the correct take to me, but I definitely have to brush up on my AI understanding. The field moves so fast I struggle to keep up, so I am having a hard time really assessing what matters.

My local barbershop has given up and instead of offering student discounts, it now offers “pensioners discount” lol.

Next step: “We’re building AGI from this bakery”

Unrelated anecdote: my iPhone tried to autocorrect “pediatrician” to “prostitution” and it still flags “pediatrician” as a “wrong word”. Get your sh*t together, @Apple …

Story Time. Two different pediatrician practices in our neighbourhood: - in the heart of my neighbourhood: one young doctor, does instagram, extremely well dressed, fancy, fashionable and “presentable” (painted the ceiling of the waiting rooms with cute children mythical figures). Very probably single, no children, talks in a way too friendly manner, like a consultant using long sentences to impress a high paying client (and you’re a high paying client, believe me 💸). Never criticises you. The assistant makes you coffee with a nice espresso machine while you wait, you can sign everything digitally… - at the border of my snob neighbourhood: one middle aged doctor, comes in with sneakers, practice is functional and quite bare, not shiny. Had 5 children during university and still got to specialise and child neurosurgery. Always got dark circles under the eyes and a coffee in the hand, talks to you in short sentences like a nice general giving orders to an inexperienced subordinate, charges legal minimum. Will tell you if you did something wrong while looking at you in the eyes, like “you must do better”. You bet your ass we’re going with the no-nonsense, 5-times-parent-while-dunking-on-university, bare-concrete-practice, doctor officer who doesn’t treat us like snowflakes.

That’s actually a very impressive efficiency in the use of compute.

People seem to think this is me “dissing” Mistral. It’s not. I’m comparing Mistral to American companies that have access to the most powerful capital-allocation machine in history and who have a head-start in the race. What Mistral achieved with those odds is impressive.

Okay so: 1) Le Chonk seems to not perform nearly as well as the models of the other frontier labs. 2) Mistral’s marketing team is outperforming all the other’s frontier labs marketing teams.

Germany has all it takes to create healthy and functional security institutions, but it’s hard, long and difficult work that needs to be supported by the right culture and governance, and that work has been neglected for decades. After 2022 I expected issues, but not this big.