Nouha Dziri

I’m a senior research scientist at Cohere Labs working with Joelle Pineau. Previously, I was a research scientist and a postdoc with Yejin Choi at Ai2 where I co-led the safety and post-training effort to build (OLMo). I got my PhD from the University of Alberta & Amii in 2023.

Currently, I’m working on advancing the science of multi-agent LLM systems.

In the past, I worked on:

• Understanding Reasoning Limits of LLMs: I study how and why LLMs succeed or fail to generalize on OOD reasoning tasks. Is it true algorithmic understanding or pattern matching? Check out my works (Faith and Fate; NeurIPS 2023), (Generative AI Paradox; ICLR 2024), (OMEGA; NeurIPS 2025)
• Advancing Reasoning Capabilities in LLMs: I develop training methods and datasets to systematically advance reasoning in mathematical and code domains. Check out the RL Grokking Recipe and my contributions to post-training large-scale efforts (Tulu 3 & OLMo2; COLM 2025).
• Ensuring LLMs Safety & Security: I ensure that safety & Security advance in step with model capabilities. I led the creation of a safety suite for OLMo, including an automated red-teaming framework (Wildteaming; NeurIPS 2024) and a safeguarding system (WildGuard; NeurIPS 2024). I also study societal risks of human over-reliance on AI, measuring reliance behaviors (RelAI, NAACL 2025) and their impact on human creativity (Artificial Hivemind; NeurIPS 2025).

My work has been featured in QuantaMagazine, TechCrunch, Science, LeMonde, Science News, etc.

I was fortunate to work with brilliant researchers in the field. I have worked with Siva Reddy at Mila/McGill, with Hannah Rashkin, Tal Linzen, David Reitter, Diyi Yang, and Tom Kwiatkowski at Google Research NYC and have worked with Alessandro Sordoni, and Goeff Gordon at Microsoft Research Montreal.