Nathan Lambert
Nathan Lambert researches the training stages that shape a language model after its initial learning. His shared work includes Zephyr, RewardBench, and Tülu 3. These projects examine preference training, reward-model tests, and model-development recipes that others can inspect. During his Ai2 period, he led post-training work; his public newsletter, Interconnects, explained methods and discussed changes in the field. His RLHF book was published in July 2026, with online material and a code companion. On 1 October 2026, Lambert and Tom Zick announced Trillium Labs. He is its co-founder and executive director. The lab plans shared post-training methods and research artifacts.
- Now at
- Trillium Labs
- Works
- 60
- Citations
- 877
- h-index
- 14
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
Public writing
Interconnects
Lambert develops a research newsletter that becomes regular after ChatGPT’s release.
-
2023
Zephyr
He joins the team studying preference distillation and direct preference optimization.
-
Ai2 period; 2024
Evaluation and model recipes
His post-training work includes RewardBench and Tülu 3 with their research teams.
-
July 2026
RLHF book
Lambert publishes an organized guide to human feedback and post-training methods.
-
October 2026
Trillium Labs
Lambert and Tom Zick announce the lab; Lambert is its executive director.
Sources and credits
Photo: Nathan Lambert personal site. License: Not stated by source. Source: natolambert.com. Cropped to a square and resized. Ask us to remove or replace this photo.