Why Open Science Is Dying at Frontier AI Labs | Nathan Lambert
MTS
October 9, 2026
Nathan Lambert announces the launch of Trillium Labs, a new nonprofit dedicated to frontier AI research and research infrastructure. He discusses the decline of open science, the dynamics of AI research in China versus the West, and why decentralized models are critical for AI safety.
Nathan Lambert
I think that all the efforts on open source have been great, and we just need more institutions and more investment in order to make science functioning as things are changing so fast. I think it's like, we've kind of gone through this regime of pre-training in the open research community. It takes months or a few years, and like, okay, I think the likes of AITU, Moran, Hugging Face have done some genuinely great open source pre-training work. We have a ladder in this methodology. And then there's kind of been the scale up in complexity of training models as we're getting multi-agent systems, very large scale RL, whatever is coming next as we're getting towards RSI. And like, this complexity hasn't kept up in the open literature.
SPEAKER_2
Hello, everyone, and welcome back to MTS. Today, we are joined by Nathan Lambert, who previously led post-training research at the Allen Institute for AI, working on open models including Olomo and Tulu, and writes the interconnects newsletter. But last week, this is the exciting news on why you're actually here at MTS. You guys officially announced the launch of Trillium Labs. This is a new nonprofit dedicated to making Frontier AI research more open, and this is really what I wanted to talk to you about today. So, Nathan, welcome to MTS.
Nathan Lambert
Thanks. Happy to be here.
SPEAKER_2
Happy to have you. I think what I'm curious about is a lot of your previous work, you've been working on open models, you were at the Allen Institute for AI.
Why, what were they all missing? Like, why did you need to start Trillium Labs?
Nathan Lambert
I think that all the efforts on open source have been great. And what, we just need more institutions and more investment in order to make science functioning as things are changing so fast. I think it's like we've kind of gone through this regime of pre-training in the open research community, takes months or a few years and like, okay, I think the likes of AITUMA and Hugging Face have done some genuinely great open source pre-training work. We have a ladder in this methodology. And then there's kind of been the scale up in complexity of training models as we're getting multi-agent systems, very large scale RL, whatever is coming next as we're getting towards RSI. This complexity hasn't kept up in the open literature.
SPEAKER_2
Yeah. It's pretty interesting because if you go back to history of just the literature behind AI, a lot of it was open. I mean, the Transformer papers kicked off modern day AI, and just a lot of our path, like it was a lot of academia, a lot of it was out in the open, everything was being built, and a little bit more of that transition, especially when most of that Intel is happening in post-training, it's happening in RL environments, it's not being told as much. And so can you just tell me that whole landscape, like let's go through the history, how to start and where are we today when it comes to just the overall openness of how we're building these things?
Nathan Lambert
Yeah, it's kind of a tricky question. I would say like the industry is so vibrant and complex because so much was open to start with. And there's just like, there was such a wide space of ideas that we're going to scale and are successful. And that's why we have so many different companies and there's kind of this like extended diaspora of the original super believers in deep learning that deserve a lot of credit for being very early on the scaling regime. And that's kind of like the earlier way of building language models before this reasoning and especially agentic capabilities took off. And like, there was enough public knowledge in this area because a lot of the methods were known. You scale them up and industrialize them.
A lot of the things that the open research does is directionally very similar in closed labs. There's just more complexity and more people working on different domains. And I think this like agentic work especially has become so infra heavy and just come at a time after so many people from academia. You go on leave, professors going on leave of apps and students dropping out of their PhD. It's just been time for this talent move to happen. So we went through this wave and then this big change in paradigm of training has happened. So I don't think it's like a super complicated history. It's just the reality of companies don't want to share because there's so much money to be made.
Read the other 22 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000793984274