Diogo Almeida
Co-Founder & Chief Scientist · Key inventor of ChatGPT/InstructGPT alignment; co-founder of TypeSafe AI; pioneer of non-autoregressive System 1 models for agent infrastructure.
Stance on AGI
"Believes that autonomous software agents will not scale if every deterministic decision relies on slow, expensive autoregressive reasoning loops. Advices dividing AI architectures into fast System 1 executors and deep System 2 reasoners."
Diogo Almeida is a machine learning researcher and co-founder of TypeSafe AI. Previously at OpenAI, he was one of the core contributors to reinforcement learning from human feedback (RLHF), InstructGPT, and early ChatGPT alignment. At TypeSafe AI, he designed Jev, a non-autoregressive System 1 model that executes structured token prediction and schema validation in sub-15 milliseconds.