Nathan Lambert Launches Nonprofit Trillium Labs for Open Post-Training Science
On October 3, AI researcher Nathan Lambert (author of the RLHF book, formerly of Hugging Face and Allen AI) announced that he is co-founding a nonprofit, Trillium Labs, with longtime collaborator Tom Zick. The news was covered by Wired, and the community response was enthusiastic.
Confirmed
- Trillium Labs is a nonprofit co-founded by Nathan Lambert and Tom Zick.
- Its first priority is building and releasing open post-training recipes.
- It plans to expand into open infrastructure.
- Research directions include RSI (recursive self-improvement), reward hacking, and multi-agent systems.
- The announcement was amplified by institutional accounts including Hugging Face, NVIDIA AI, Ollama, and TheTuringPost.
Why it matters
- Post-training recipes are a key differentiator of frontier model competitiveness, and few details are public; sustained open releases could significantly lower the barrier to reproducing frontier capabilities.
- Lambert has long shared post-training insights publicly; institutionalizing this work as a nonprofit may become a key supply point for the open-source community in the post-training era.
2026-10-03 ~ 2026-10-04 · 26 related posts
Primary sources
- Nat Lambert co-founds Trillium Labs, a nonprofit for open frontier AI science — natolambert ·
- Nathan Lambert Launches Trillium Labs, a Nonprofit for Open Research on RSI and Reward Hacking — natolambert ·
- Nathan Lambert Launches Nonprofit Trillium Labs for Open Post-Training Recipes and Frontier AI Science — reach_vb ·
- [source] Nat Lambert co-founds Trillium Labs, a nonprofit for open frontier AI science — natolambert · 2026-10-03
- [source] Nathan Lambert Launches Trillium Labs, a Nonprofit for Open Research on RSI and Reward Hacking — natolambert · 2026-10-03
- Nathan Lambert launches Trillium Labs, a nonprofit for open frontier AI science — himanshustwts · 2026-10-03
23 near-duplicate retellings: ricklamers · ollama · morgymcg · samsja19 · suzatweet · huggingface · CFGeek · giffmana · TheTuringPost · sethlazar · tokenbender · NVIDIAAI · divy93t · MaziyarPanahi · reach_vb · omarsar0 · aidenybai · maximelabonne · zacharynado · shaneguML · HannaHajishirzi · evijit · deliprao