A steal of a deal for the impulsive among you: Sign up for my Data Art In The Jungle™ retreat which starts in 13 days and I'll give you free registration to the autumn Binoculars to Binomials course! A $645 value! www.jerthorp.me/btb-retreat-... Flights are super affordable to Colombia right now!

I feels to weird to tell LLMs not to do things "by hand"

Do I get tired of being so right with LLMs? Yes Is it worth it? Definitely The time gains to think properly about problems, and the best solutions to a given situation is simply incredible

Programming as the creation of any other communication codes is the art of leaving things out

✨ We're thrilled to announce our new membership director: Elaine Diaz Rodriguez! She will run the membership experience, including training, cohort meetings and more. She previously helped lead coaching at @lionpubs.bsky.social. Welcome to the team, Elaine! 💖 www.tinynewsco.org/announcing-t...

Over the next couple of weeks we will be stitching together how we do data processing simple for small data: spreadsheets, maps, charts, data transformations

Not everybody achieves mastery, mostly because they get tired of doing the same thing over and over. Rewrite, refactor, redraw, again and again, from scratch, many times over... feels like time wasted, but it is not.

Progress is like entropy... there is no going back

The Pareto Law of Vibe Coding: AI can get you to 80% of a product quickly, but the final 20% takes 80% of your time. Real-world use cases require systems thinking and problem decomposition — something non-coders will struggle with.

AI is bringing us closer to leverage what unit testing did for software but for public policy

Many people say that we will be replaced not by AI but by people who knows how to use AI, which I think it is true. But there is something deeper, I believe the professionals who will thrive in the AI era are actually systems thinkers with good taste, good aesthetics

It took me a month to code a project (40k lines of code)... something that otherwise would have taken me a year or two Thanks to AI and Cursor

Many of the worlds problems are problems of jack plugs of information flows

I am creating a system in typescript with 1-index based arrays

The world moved from Swordsmiths to Codesmiths... AI tools and agents orchestration will be more a craft that an science. AI Craftsmanship will change education and first time jobs as we know them. No degree required, learn with the masters.

This is just crazy... I plan with cursor a big refactor and translation from one code base in one language to another language. The plan results in a timeline of weeks, as the LLMs guesstimate from human times. But the actual migration will go through in a couple of hours instead

You don't know exactly what you need to do, until you do

Mosquito holidays

Is there some type of epistemological complexity metric? Kolmogorov complexity won't work for optimizing or simplifying LLM written code. Will verbosity and explainability reign over clever code?

Vibe coding be like

It is very difficult to make things simple, as it requires a great level of complexity under the hood, which may feel counter intuitive. The way the complexity is handled by little interconnected blocks gives rise to beauty, and beauty is the best assessment of simplicity

Chatgpt started referring traffic to @datasketch 😲

If you are not getting "I see the issue" at least 10 times a day, you are not building enough

Had fun catching up with some coding... ended up building a bunch of base packages to help me work with images and spreadsheets. Here is a preview of what I did, but first checkout this Mona Lisa

I got the flu so couldn't work for a couple of days this week, time to get back on track. Test: how many small data extensions can I code in a couple of hours?

Friday night vibe coding... CamFilter just saw the light... magenta by default From cero to working app 1 hour. Fixing the details 3 hours more :) Happy with the results and the selfies from the laptop

“Everybody in math and physics uses it.” Nearly 35 years ago, Paul Ginsparg created arXiv, a digital repository for sharing research. It changed science for good. Three decades in, Ginsparg reflects on its legacy—and its future.

Yesterday I vibe coded from scratch in 4 hours an R package that took me 4 weeks to build a while back

How would you feel about a book that would rewrite the upcoming pages as you read it through?

Just got to NYC. I will be participating in the Open Data Week. It all started with a question: Why is it difficult to get data about Spanish readers or the publishing industry in Spanish? Here is preview of the timeline of some Latin American writers in NYC

I've been using google less and less... but after a few hours of fighting cursor, deno, node, duckdb, CommonJS, ESM... the answer was right there: as one of the firsts google results in a github issue

Day 15 Signed up our first 2 partners, to retry our dashboard and AI agent for data analysis, 98 to go. I've spent most of the time making demos and improving our internal workflows. 30 potentials identified, 5 proposals sent. This week need to improve onboarding and our sites

Are you left or right handed? ChatGPT vs Claude

Culture is society's persistent data storage

Day 5 5 new leads in media, finance, environment, small businesses, academia They all need sensible data architectures and easy to use data interfaces

Day 4 Out of 14k emails sent in the newsletter got 2 replies to inquire about our solutions. Spent all day organizing LinkedIn contacts. Maybe doing something more targeted will work?

Bluesky CDMX

Day 3 2 sales calls. Increasing interest in our AI agent that brings data driven answers to any database

Day 2 Made a roadmap of extensions, small data apps to simplify data analysis inside Datasketch. Started coding our Machine Intelligence Query Question and Answer bot. Of course, had to buy a new domain.

Day 1 Let's see how we do with out 100 partners campaign. I am looking for organizations that want to leverage their own data with our AI tools Campaign page live, newsletter, sent. I'll post daily updates on how we can reach and find new partners

The unification of physics will come when we type down the formula for energy ~ information

Memory makes computation universal, remember? thinks.lol/2025/01/what...

People should not learn AI People should learn to learn with AI

When I started my company, I wanted to hire myself. Mostly but because I knew exactly what I needed... I needed someone to execute my vision, but that vision is hard to convey

I learn a lot by working with interdisciplinary teams... the craziest part... now I can see an AI with any all backgrounds work and see how they solve problems in their own domain Infinite inspiration!

This is how LLM driven AI feels to me... like "trying to kill a fly with a bazooka"

As an entrepreneur, one of the biggest challenges is coping with having so much on your plate. One thing that has helped me is focusing not on what I should do, but rather on what I shouldn't do... Enter the OMIT framework...

We can have a team of PhD level agents working for us with advanced reasoning... But here is the catch, are you able to manage PhD students? And if you do, are you really solving PhD level stuff in your day to day?

If anything AI will make us more human

And there’s one thing: users of data are not only humans—LLMs are also users of data. However, as of now, even though LLMs can run Python and build JavaScript artifacts, they are far from being usable in general settings due to data volume (token limits) or lack of context