I declare myself allergic to self-proclaimed experts. www.pelayoarbues.com/notes/Self-p...

Once again I really, really needed this for a better Friday! It is more relevant and hilarious every time I visit. alignmentalignment.ai

Just finished reading 3 Millones de Viviendas by @jorgegalindo.bsky.social. One of the sharpest takes on Spain’s housing crisis I’ve seen. Data, history, and a tough question: what kind of country do we want to build? www.pelayoarbues.com/notes/Tres-M... #realestate

I'm ready to accept a definition of "agent" that I think is widely-enough agreed upon to be useful, at least in the AI engineering space: An LLM agent runs tools in a loop to achieve a goal (I've been dismissing the term as hopelessly ambiguous for years) simonwillison.net/2025/Sep/18/...

Context engineering is the new AI buzzword. And here we go again: www.pelayoarbues.com/notes/Contex...

The data industry promotes builders. Not leaders. You nailed pipelines. So they gave you politics. Now you're managing people, budgets, and roadmaps. With zero training. You’re not unqualified. You’re just undertrained.

If you load this page it contacts 82 IP addresses executing 256 separate HTTP transactions to download 18MB of data writing 64 cookies to your device to tell you “no”

OpenAI announced Codex (openai.com/index/introd...) on May 16, 2025 for their $200/month Pro users, and Google announced a competing tool called Jules (jules.google) on May 20, 2025 making it available for FREE. There’s no moat in AI—how can a moat disappear in less than a week?

Veo 3: "a big broadway musical about garlic bread, with elaborate costumes and a sondheim-like vibe"

Every AI coding tools company doing demos with React + Typescript (and maybe React Native.) Microsoft meanwhile showing off Java and .NET case studies. This doesn’t look like an accident - but a very deliberate strategy. It’s easier to win when you are on a ground with fewer competitors…

VEO 3 Prompt: Generate a TV news anchor with a British accent, the anchor says "in shocking news, Jake Kaye Rowe ling's yacht sank with her on board after being attacked by orcas off the coast of turkey" include the announcement in the dialogue. This took one prompt and 2 minutes to generate.

500 days reviewing highlights on @readwise.bsky.social!!!

I just shared a learning path guide for software developers looking to infuse AI into their products, tailored specifically to your needs, interests, and how deep you'd like to dive. www.pelayoarbues.com/notes/AI-Lea...

When people are pressured to meet a target value there are three ways they can proceed: 1) They can work to improve the system 2) They can distort the system 3) Or they can distort the data — Donald Wheeler commoncog.com/goodharts-la...

I don't know why people think picking a model on ChatGPT is so complicated.

SmolVLM real-time camera is a framework for running computer vision models locally on M3 MacBooks. It combines HTML webcam access with llama.cpp's inference process to enable real-time image capture, understanding, and captioning. www.threads.com/@luokai/post...

It is a little weird to me countries aren’t more aggressively, formally trying to take advantage of the U.S. science brain drain. Once in a lifetime opportunity to buy low on Non-Dumbass Americans with PhDs who just wanna look into microscopes and quietly cure ass cancer as our country eats shit.

Nuevo post. Área de aplicación. Cosas curiosas que he leído este finde y que me parecen interesantes. Reproduzco parte de un post de Jakub Nowosad muestrear-no-es-pecado.es/2025/05/aoa

Just finished 'Power and Prediction'. Loved the economic take: AI dramatically reduces prediction costs but leaves human judgment scarce. Also, great analogy with electricity, real AI impact requires systemic change www.pelayoarbues.com/notes/Power-...

Spain's power is slowly coming back, but situation still far from normal. Out of the 33 GW scheduled for this evening peak, only 18 GW is available. Nuclear and coal-fired power plants still offline.

When we don’t know someone well, we tend to project intentions onto them, casting them as a character in a play they themselves are unaware exists.

Why software engineers should learn a bit of data science www.pelayoarbues.com/notes/Why-So... #databs

In simple terms, the purpose of data is to give you a causal model of your business in your head. If reality changes, the figures in your business should show it. At which point you get to update your knowledge.

Coaching is no longer a specialty; you cannot be a good manager without being a good coach. The path to success in a fast- moving, highly competitive, technology- driven business world is to form high- performing teams and give them the resources and freedom to do great things.

When leaders take steps to protect their own interests, particularly when those steps are taken at the expense of others, they send a message to everyone else that it is okay to do the same.

To maintain the core of your product there are usually one or two things that have to stay still while everything else spins and changes around them. And that’s a useful constraint. You need some constraints to force you to dig deep and get creative.

Yes, leadership is about vision. But leadership is equally about creating a climate where the truth is heard and the brutal facts confronted

If people only comply, we can’t expect people to take responsibility for their actions. The chain of command is for orders, not information. Responsibility is not doing as we are told, that’s obedience. Responsibility is doing what is right.

The push for commercial control of AI capability is dangerous. Naomi Klein, who coined the term “shock doctrine” as “the brutal tactic of using the public’s disorientation following a collective shock… to push through radical pro-corporate measures”

What AI person is saying..... Anthropic CEO, Dario Amodei: In the next 3 to 6 months, AI is writing 90% of the code, and in 12 months, nearly all code may be generated by AI. What AI person is doing..... www.anthropic.com/jobs

Having two different pipelines to process your data is a common cause for bugs in ML production. One cause for bugs is when the changes in one pipeline aren’t correctly replicated in the other, leading to two pipelines extracting two different sets of features.

Every three to six months, audit your meetings. Are they useful? Are the right people involved? Do they achieve their purpose, and are they worth the time? Could more be accomplished asynchronously? Consider polling participants or having a neutral person do so.

Not finance. Not strategy. Not technology. It is teamwork that remains the ultimate competitive advantage, both because it is so powerful and so rare.

Make sure your candidates know that you’re excited about them. I first encountered this idea reading Rands’s “Wanted” article, 4 and he does an excellent job of covering it there. The remarkable thing is how few companies and teams do this intentionally.

Self-awareness has three components: understanding your underlying value system, identifying your innate preferences—your work style and decision-making tendencies—and being clear about your own skills and capability gaps.

To write something, you have to do things, collect ideas, process them, put words on the page, structure them, edit them, rewrite them, and publish them to your site — and that’s all for one article. Then, there’s the emotional challenges: writer’s block, imposter syndrome (...)

Tenéis que leer esta reseña del concierto de Duran Duran dirigido por David Lynch (?!).

Software engineer job postings are at a 5-year low, as per data from Indeed (which is a pretty good data source). But how bad are things, really? I looked closer and found that it's... probably not as bad as circulated on social media (cont'd) Also: no, it's not the end of software engineering...

"Any given executive is almost always uncannily good at one way of consuming information. They feel most comfortable consuming data in that particular way, and the communication systems surrounding them are optimized to communicate with them in that one way." By @lethain.com

I would argue operational clarity is exactly the opportunity many data teams have been looking for. So many leaders we speak to say something like “We can do so much more.” Adding operational clarity is something undeniably valuable that data folks are uniquely positioned to do well.

Gobierno y oposición acuerdan empezar una guerra civil para volver a librarse de la Guerra Mundial. www.elmundotoday.com/2025/02/gobi...

Better products, services and experiences are usually the result of the employees who invented, innovated or supplied them. As soon as people are put second on the priority list, differentiation gives way to commoditization. And when that happens, innovation declines (...)

I once read somewhere on the internet: deploying is easy if you ignore all the hard parts

Being able to set ego aside and be ruthlessly critical of yourself and your division is hard. It’s not just about where you have gaps today but also about what you’ll need in the next three to five years.

TFW you realize how many hundreds of billions have been made by first convincing people that structured data stores are too hard to set up and maintain, then 5-10 years later convincing the same people that unstructured data stores are not conducive to fast and reliable analysis. Rinse, Repeat.

"You can’t manage what you can’t measure” is a maxim that is taught and believed by many in both the business and education sectors. But in fact, the phrase is ridiculous. A large portion of what we manage can’t be measured, and not realizing this has unintended consequences

The growth in building footprint area between 1950 and 2020 in Spain is impressive. The dataset was published in ESSD, #HISDACSpain 100 x 100 m resolution is available for free use. More information can be found here 🧑‍💻 essd.copernicus.org/articles/15/... #OpenAccess #OpenData #geography

We need to be tough in the tech industry, but we don’t have to be cynical all the time. It's important to celebrate those who create ideas and shape the future by influencing the narrative in a healthy way. I'm grateful for people like Chip Huyen in this industry.

Just tested Mistral's Le Chat, works really well and super fast. This prompt did a great job: chat.mistral.ai/chat.

This paper is wild - a Stanford team shows the simplest way to make an open LLM into a reasoning model They used just 1,000 carefully curated reasoning examples & a trick where if the model tries to stop thinking, they append "Wait" to force it to continue. Near o1 at math. arxiv.org/pdf/2501.19393