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.

FlexOlmo: Federated Learning for LLMs?

What if you could train an AI model with data from several companies, without any of them sharing their sensitive data? That's the idea behind Ai2's FlexOlmo.

Kimi K2: the new SOTA open source AI model is Chinese!

Kimi K2, from the Chinese company Moonshot AI, has been released

Energy-Based Transformers: Toward AI Models That Know When They Don't Know

A new path to teach AIs to think slowly

SEAL: AIs That Learn to Learn

What if LLMs could "learn to learn"? A new MIT paper, Self-Adapting Language Models (SEAL), lets a model generate its own training data and fine-tune itself.

Guide: How to Use AI Coding Agents

Cognition, the company behind Devin, released a guide of best practices for coding agents: treat them like a teammate, not an autocomplete.

Fast and Furious: Diffusion Models Generating Text

Inception AI Labs launches Mercury, an extremely fast model

Anthropic Teaches How to Build AI Agents

Anthropic opened the hood on the multi-agent system behind Claude's Research feature and shared the lessons it learned along the way.

Is There Originality in AIs?

I gave the most powerful AIs the same zero-shot request: an original sentence so brilliant it deserves an article. Here's what each one answered.

The frantic pace of AI releases

Do you have any idea how many AI models (and versions) were released in 2025 alone? Here's a month-by-month summary.

The Illusion of Reasoning in AIs

An Apple paper questions whether AIs really think, or just pretend to