A group of 25 Fields Medalists, including myself, have made a joint declaration on Math and AI: mathandai.org . We welcome additional signatories. See also this article in the Economist announcing the declaration: www.economist.com/science-and-...

Abrimos posición en el equipo de datos de idealista/data! Un rol muy de remangarse, hacer nuevas versiones de productos de datos existentes, trabajar con el equipo de negocio, ayudar a migrar a nuevo stack y mucho ELT. www.linkedin.com/jobs/view/44...

what’s your favorite way to have your data? data base data center data mart data warehouse data lake data stream data pipeline data mesh data fabric data swamp data silo data cube data frame data vault data catalog data platform data store data pipe data table data feed data channel data bus

Save your data to local Markdown files you own forever, and can access completely offline. Obsidian now helps you migrate from all the major proprietary formats including Notion, Evernote, OneNote, Google Keep, Apple Notes, and more.

So we know AI helped a negligible number of students. But how many did it harm? I've crunched the raw data provided by the lead researcher, and what I've found is that ~9,500 students using AI studied less than even the *least* studious, non-AI using student. This should shock the conscience.

Benjamin Riley@benjaminjriley.bsky.social · last mo.

Both the Economist and others suggest this study demonstrates that AI can help students. So I contacted the lead researcher to figure out how many. Answer: Out of ~26,000 kids, a total of 20. And five months after starting to use AI, only four continued to spend 65 minutes on homework. Four.

We lost reading. www.pelayoarbues.com/notes/we-los...

We lost reading

We lost reading. In the advent of GenAI we are all worried about losing the writing. Software developers write less code than ever.

pelayoarbues.com

Autonomous AI in action. 🤖 Check out how the new Gemma 4 31B model operates as an ADK Agent, exploring, planning, and running experiments on an unfamiliar database to optimize services and maximize revenue. 📈 ✨ Dive into the full session here to explore the complete workflow: goo.gle/4ozP8sK

When people say 'AI isn't a bubble because the technology isn't going away', it just shows they don't know what a bubble is. The internet didn't go away. Railways didn't go away. Tulips didn't go away. An asset can be overvalued, even if its useful.

Mamá salgo en @forbes.com! Podeis escuchar/ver el podcast con la entrevista que me hicieron en Open C. forbes.es/brandvoice/9...

Performanze lanza su nuevo podcast con Pelayo González Arbués como primer invitado: «La gente espera magia de la IA, y eso rara vez ocurre» - Forbes España

La consultora tecnológica presenta 'Open C', un espacio de conversación con directivos sobre cómo cambian las decisiones en las empresas con la tecnología presente. En su estreno, el director de Cienc...

forbes.es

I have been working on Obsidian Reader for over a year. I didn't want to share it until I felt it was good enough. It's finally there. Consistent formatting for any article. Outline, syntax highlighting, nice footnotes, adjustable typography. Runs locally. Just rules, no AI.

#StarterPack #Saturday @pelayoarbues.com put it all the way together right here! If you are like #MissKitty pulling levers with #AI #Research, then you want this. If you want to get your two cents in. That is there. Come on and get a #FollowAll and #Repost and listen in or tell it all!

When you search for a home, it’s rarely “the whole city.” It’s more like: this neighborhood yes, this one no, that other one too, even if it’s far away. With @idealista.bsky.social new multi-area search, you can mix, exclude and add zones in one map search.

I am calling for a complete and total boycott of the Mercator projection in all news stories about Greenland until every member of the American public has seen this

The books I read in 2025. I followed the same habit that worked so well in 2024: reading non-fiction first to satisfy my curiosity and then switching to fiction to wind down. This routine is still the best compromise I’ve found between learning and resting. www.pelayoarbues.com/notes/Wrappe...

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

Center for the Alignment of AI Alignment Centers

We align the aligners

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?

Jules - An Asynchronous Coding Agent

Jules is an asynchronous agent that gets out of your way. It lets you focus on the coding you want to do, meawnwhile picking up all the other random tasks that you rather not do.

jules.google

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...

Goodhart's Law Isn't as Useful as You Might Think

Goodhart's Law is useless. It tells you about a phenomenon, but it doesn't tell you how to solve it. We look at how organisations actually prevent Goodhart's Law, and illustrate this with Amazon's Wee...

commoncog.com

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...

luokai (@luokai) on Threads

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, ...

threads.com