# Nathan Lambert

> Nathan Lambert is a Senior Research Scientist and Post-Training Lead at the Allen Institute for AI (Ai2), where he leads open-source language model development on the OLMo and Tulu series. A UC Berkeley PhD, he previously led the RLHF team at Hugging Face, co-building the TRL library and the Zephyr model. He runs Interconnects AI, a Substack newsletter read by tens of thousands covering post-training, open models, and AI policy, and is the author of The RLHF Book (Manning Publications). With roughly 8,000 academic citations and a reputation for demystifying the hardest parts of modern AI, Lambert is one of the most trusted voices at the intersection of open-source AI research and public education.

- **Role:** Senior Research Scientist and Post-Training Lead at Allen Institute for AI (Ai2)
- **Organizations:** Allen Institute for AI (Ai2), Interconnects AI, SAIL Media
- **Nationality:** American
- **Education:** PhD, Electrical Engineering and Computer Sciences, UC Berkeley
- **Known for:** Co-built TRL (Transformer Reinforcement Learning), one of the most widely used open-source RLHF libraries, Released Zephyr-Beta, an early influential RLHF-trained open model, Co-authored Tulu 2 - first model demonstrating DPO scales to 70B parameters

## Career timeline

- **2017** — Began PhD at UC Berkeley, EECS department; advisors Kristofer Pister and Roberto Calandra
- **2019** — Interned at Facebook AI Research (FAIR), recruited by Roberto Calandra; formative experience in running AI experiments at scale
- **2020-2021** — Interned at DeepMind, working on model-based reinforcement learning
- **2022** — Completed PhD; joined Hugging Face as Lead Scientist on the RLHF team, recruited by Douwe Kiela
- **2023** — Co-built TRL (Transformer Reinforcement Learning) library and released Zephyr-Beta model; began scaling Interconnects AI newsletter post-ChatGPT
- **2024** — Joined Allen Institute for AI (Ai2) as Senior Research Scientist and Post-Training Lead; led Tulu 3 and OLMo 2 releases
- **2025** — Led OLMo 3 32B release (best open-source model at release); Interconnects AI hit 3.5M+ page-views; The RLHF Book entered Manning Early Access
- **2026** — Appeared on Lex Fridman Podcast; RLHF Book ArXiv paper companion published (arxiv.org/abs/2504.12501)

## Achievements

- Co-built TRL (Transformer Reinforcement Learning), one of the most widely used open-source RLHF libraries
- Released Zephyr-Beta, an early influential RLHF-trained open model
- Co-authored Tulu 2 - first model demonstrating DPO scales to 70B parameters
- Led post-training for OLMo 2 (7B, 13B, 32B) and OLMo 3 series at Ai2
- Grew Interconnects AI to over 3.5 million page-views in 2025
- Approximately 8,000 academic citations for AI research
- Authored The RLHF Book (Manning Publications, 2026)
- Co-leads Open Instruct, Ai2's open post-training codebase
- Over 300 newsletter posts published on Interconnects AI

## Latest updates

- **2026-04** — ArXiv companion paper for The RLHF Book published (arxiv.org/abs/2504.12501)
- **2026-02** — Appeared on Lex Fridman Podcast covering the full landscape of AI
- **2026-02** — Interview with Turing Post on open models, reasoning systems, and where AI is headed
- **2025-12** — OLMo 3 32B released, described as the best open-source language model at time of release
- **2025-00** — Interconnects AI newsletter surpassed 3.5 million page-views for 2025
- **2025-00** — The RLHF Book entered Manning Early Access Program (MEAP)

## Links

- Website: https://natolambert.com
- LinkedIn: https://www.linkedin.com/in/natolambert/
- Twitter/X: https://x.com/natolambert
- GitHub: https://github.com/natolambert

---

Profile page: https://yespress.io/nathan-lambert
Published by YesPress — https://yespress.io
Last updated: 2026-04-24