ES EN

My tweets from April 2026

2026-04-30

April 22, 2026

The more I think and read about consciousness, the closer I get to panpsychism.

❄ ❄ ❄ ❄ ❄

It is amazing what the new version of OpenAI’s image model can do. I gave ChatGPT the description of my docflow repo and it generated this.

https://github.com/domingogallardo/docflow

❄ ❄ ❄ ❄ ❄

In 13 minutes Codex changed more than 800 lines of code across 13 files.

It expanded an existing feature, generalizing and adapting it: before, Docflow only downloaded my liked tweets. Now it also downloads my posted tweets.

https://github.com/domingogallardo/docflow/commit/0a62a8a37bf55c77d3c7c8a26a7e15079460178d

❄ ❄ ❄ ❄ ❄

Listening to Patel’s podcast with Jensen Huang, I was struck by how Anthropic-pilled Patel is. On chip restrictions to China, Huang’s collaborative stance seems far more reasonable to me than Patel/Anthropic’s fear-based stance.

April 23, 2026

I ran an experiment with GPT Image 2: I gave it 5 photos from a sequence and asked for the next one.

The result shows that OpenAI’s new model is capable of preserving the spatial logic of the scene and generating a coherent view, very much in the spirit of DeepMind’s Genie.

❄ ❄ ❄ ❄ ❄

I gave GPT Image 2 five photos from a sequence and asked for the sixth.

What came back wasn’t just a plausible image. It was a coherent next view.

Spatial consistency, camera progression, object continuity.

April 24, 2026

Infographic to put Epoch AI’s post about the available power at OpenAI’s Stargate Abilene into context.

❄ ❄ ❄ ❄ ❄

Well, it does work for me in GPT 5.5.

April 25, 2026

Four months. We are in the singularity.

April 27, 2026

In case anyone wants to try @Jongonzlz’s code from the command line, I created a repo you can download.

The CLI instructions:

https://github.com/domingogallardo/calculadora-salarios-irpf-cli-demo/blob/main/CLI.md

And the original addition to his repo is here:

https://github.com/jongonzlz/Calculadora-de-Salarios-y-Progresividad-en-Fr-o/pull/4

April 28, 2026

I published the repo with the instructions for the AI tutors we use in the University of Alicante course Programming Languages and Paradigms.

Copy it if you want your own tutor.

❄ ❄ ❄ ❄ ❄

“The internal representations [of LLMs] are shaped by linguistic data that reflect patterns or structures of reality […]. One could say they have crossed the threshold of meaning. […] That does not mean they have a mind or, much less, consciousness.”

https://vonneumannmachine.wordpress.com/2026/04/28/the-vector-grounding-problem-comprenden-algo-los-llms/

April 29, 2026

This tweet has gone viral. I looked at the company’s website and it seems serious, with competent people behind it and interesting scientific publications. Are we looking at something new and relevant? I do not know. Just in case, I have already signed up for the beta.

❄ ❄ ❄ ❄ ❄

Watch the EPA employment data for programming and technology. The negative impact of systems like Codex and Claude Code is starting to show up.

Will there be a Jevons effect? My feeling is no: a small number of engineers will be able to orchestrate all the enterprise software that needs to be produced. The rest will be done by users interacting with their personal agents.

April 30, 2026

Life keeps getting easier for product owners and product managers.

❄ ❄ ❄ ❄ ❄

More GPT Image 2, this time with a two-page spread about Alan Kay’s Dynabook. Made with AI.

❄ ❄ ❄ ❄ ❄

New interview/podcast with Demis Hassabis. One of the most interesting ones he has done recently, with important predictions and insights:

  • AGI around 2030
  • A complete virtual cell in about 10 years, that is, around 2036
  • Full value from agent tools / vibe coding within 6-12 months from April 2026
  • Frontier capabilities distilled into small models within 6-12 months after appearing in large models
  • Solving Millennium Prize-type problems perhaps within a couple of years