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.

SOTAQUE: Our Open Source Dataset of Brazilian Voices
A new open source project to build an open dataset of Brazilian Portuguese voices and help AI understand and speak our accents better.

Harness: the word of 2026 in AI
The secret behind Claude Code and Cursor

Google discovers a prompt engineering trick to improve LLM answers
A Google Research paper found that simply pasting your prompt twice improves the performance of non-reasoning LLMs.

Anthropic announces Claude Mythos Preview
The model that won't be released

Anthropic research on emotions in AI models
Anthropic found activation patterns in Claude that match human emotional concepts like fear, calm and desperation, and that directly change the model's decisions.

Loss Function: How AI Learns from Its Own Mistakes
A walkthrough of how LLMs actually learn under the hood during training, from the Loss Function and cross-entropy to backpropagation.

Google Embedding 2 - Complete Guide
Gemini Embedding 2 maps text, images, audio, video and PDFs into the same vector space, and my friend Pedro Gengo and I built a hands-on Colab guide to try it all.

Logits: What Happens Before the AI Picks the Next Word
A more technical look at how LLMs work under the hood: what logits are and how they turn into the next token the model picks.

Official Guide: How to Use Skills in Claude Code
Anthropic released a complete guide to building Skills for Claude Code: what a skill is, how Progressive Disclosure works and how it relates to MCP.

SkillsBench: how you should use Skills in Claude Code
A new paper tested AI agents with and without Skills across 80+ tasks, and the results say a lot about how to use them in Claude Code.