My research is generally focused on making modern language models broadly useful and reliable. I've recently been focused on improving model capabilities through post-training by creating new datasets and evaluations, and improving training algorithms and model architectures. I'm also a strong supporter of open science, and I've contributed to openly released artefacts including Tulu 3, RewardBench, Dolma, and OLMo, OLMo 2, and OLMoE, with more coming soon.