Blog posts in English
I write my posts in Portuguese first, for a Brazilian audience, and then translate them. These are all of them in English, newest first. The original Portuguese blog has the same posts.

Embeddings: The Map That Teaches AI to Understand the World
Embeddings turn text, images and audio into lists of numbers that form a map of meaning. Here is how they work, how LLMs learn them, and why they power RAG.

The Claude Fable 5 shutdown, AI sovereignty and national security
After the US government cut off Claude Fable 5 for non-Americans overnight, what does that mean for countries running critical infrastructure on third-party models via API?

Claude Fable and Claude Mythos: everything you need to know
Anthropic finally released Claude Mythos 5 and Claude Fable 5. I got to see the dragon, but through glass, with warning labels and a few asterisks.

Einstein, Douglas Adams, Arrival, and Scientific AI
Lately, my reading on the subway and before bed has been “Story of Your Life”, the Ted Chiang short story that inspired “Arrival”, one of my favorite movies because it brings together science fiction, linguistics, and communication between different species.

Anthropic says HTML is better than Markdown for AI outputs
A post by a Claude Code engineer made a lot of noise in recent weeks

RecursiveMAS: AI agents communicating by telepathy
A new paper from Stanford, MIT and NVIDIA proposes that multi-agent systems stop trading walls of text and start communicating directly in the models' latent space.

Harness: the word of 2026 (and how Codex works under the hood)
Why Codex and Claude Code do so much more than the same model in a chat window: what a harness is, and what I found by reading Codex's open source code.

TUTORIAL: A guide to quantizing Hugging Face AI models to GGUF and running them with Ollama
How to shrink an open source Hugging Face model with quantization, convert it to GGUF with llama.cpp, and run it on your own laptop or phone with Ollama.

With AI, I became a hostage of my own ideas
When every idea seems doable, the hard part stops being execution. It becomes knowing when to stop.

Claude Code, Doubled!
Anthropic teams up with SpaceX