I'm Martin, an AI researcher at AE Studio working on AI alignment.

• 🛡️ I work on training-time safety techniques for LLMs, with the goal of reducing existential risk from advanced AI.
• 🎓 I hold a PhD (and a Master's) in Computer Science from PUCRS, where I researched text-guided image editing and test-time compute scaling.
• 💼 Before AE Studio, I spent five years as a technical lead building LLM systems in industry.
• 🤝 Always happy to talk about AI safety research; feel free to reach out!

I've been working with deep learning since 2016. My earlier research focused on generative models and multimodal learning, especially text-guided image editing. Today my focus is AI alignment: keeping increasingly capable AI systems safe and under meaningful human oversight, including through the point where they begin to recursively self-improve.

I'm currently exploring:

• Mechanistic interpretability
• Safely scaling reinforcement learning with verifiably aligned rewards