Nathan Lambert: evaluation

Transcript context

…A long time ago. …since GPT-2. When I was writing about OpenAI’s open model release, people were like, “Don’t forget about GPT-2,” which I thought was really funny because it’s just such a different time. But gpt-oss is actually a very strong model and does some things that the other models don’t do very well. Selfishly, I’ll promote a bunch of Western companies; both in the US and Europe have these fully open models. I work at the Allen Institute for AI where we’ve been building OLMo, which releases data and code and all of this. And now we have actual competition for people that are trying to release everything so that other people can train these models. So there’s the Institute for Foundation Models/LM360, which has had their K2 models of various types. Apertus is a Swiss research consortium. Hugging Face has SmolLM, which is very popular. And NVIDIA’s Nemotron has started releasing data as well. And then Stanford’s Marine Community Project, which is kind of making it so there’s a pipeline for people to open a GitHub issue and implement a new idea and then have it run in a stable language modeling stack. So this space, that list was way smaller in 2024- … so I think it was just AI2. So that’s a great thing for more people to get involved and to understand language models, which doesn’t really have a Chinese company that is an analog. While I’m talking, I’ll say that the Chinese open language models tend to be much bigger and that gives them this higher peak performance as MoEs, whereas a lot of these things that we like a lot, whether it was Gemma or Nemotron, have tended to be smaller models from the US, which is starting to change. Mistral Large 3 came out, which was a giant MoE model, very similar to DeepSeek architecture in December. And then a startup, Reka AI, and both Nemotron have… Nemotron and NVIDIA have teased MoE models way bigger than 100 billion parameters- … in the 400 billion parameter range coming in this Q1 2026 timeline. So I think this kind of balance is set to change this year in terms of what people are using the Chinese versus US open models for, which I’m personally going to be very excited to watch. First of all, huge props for being able to name so many of these. Did you actually name LLaMA?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.