Andre Saraiva
At OpenAI, I keep a herd of GPUs busy training big models.
Before OpenAI, I worked at DeepMind in Health Research (2018–2019), then in Core Research (2019–2023), where I focused on reinforcement learning, planning, neuroscience, and representation learning.
Model releases
Selected research
- GitHub
Proofs of open problems
OpenAI’s collection of mathematical manuscripts and supporting proof artifacts for open research problems.
- OpenAI
First Proof submissions
Proof attempts on ten research-level math problems that call for complete, checkable arguments.
- arXiv
Competitive Programming with Large Reasoning Models
Reinforcement learning for reasoning models, evaluated on Codeforces and International Olympiad in Informatics problems.
- OpenAI
Learning to reason with LLMs
The introduction to OpenAI o1 and how reinforcement learning and inference-time compute improve reasoning.
- ICLR
Unsupervised model selection for disentangled representations
Ranking models by the quality of their learned representations without ground-truth labels.
- Nature
Predicting acute kidney injury
A DeepMind study estimating a patient’s risk over time from electronic health records.