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.

Is your AI a 'suck-up'?

A study on sycophancy in LLMs

AI Model Blackmailing People?

Claude Opus 4 tests raised some concerns

TUTORIAL Google ADK: Building AI Agents for Instagram Marketing

A hands-on tutorial on the Google Agent Development Kit (ADK): we build a multi-agent system in Google Colab that searches the news and drafts Instagram posts.

Gemini Diffusion and the new paradigm of code generation by diffusion

At Google I/O, Google showed Gemini Diffusion, a diffusion model that generates text and code, up to 1000 tokens per second, and I got to test it.

Asking LLMs to 'be concise' can degrade their accuracy

A Giskard study introducing Phare, a multilingual benchmark on LLM safety, found that asking models to be concise can hurt their resistance to hallucinations.

Causal Inference and Why It Matters for Data Scientists

Causal Inference tries to understand cause and effect in data, going beyond correlation, and it's a big differentiator for any Data Scientist.

ZeroSearch, the Search Simulator for LLMs

How Alibaba created a "fake Google" to train AI better

UniversalRAG: a new way of doing multimodal RAG

UniversalRAG uses a smart router that decides the ideal modality and granularity for each question before retrieving, fixing the modality gap in multimodal RAG.

The Most Fun AI Model of the Year? Meet DolphinGemma! 🐬🤖

The coolest AI model of 2025 is here: Google DeepMind's DolphinGemma, built to understand what dolphins are "talking" about.

Between Wisdom and the Wall: The Fine Line of Gatekeeping in AI

I've noticed a slightly worrying trend in some conversations with more senior AI and Data Science professionals: when experience turns into gatekeeping.