Nathan Lambert
speaker
1,814 appearances
3 recordings
2 series
first heard Feb 2025
last heard 1 Feb
Nathan Lambert’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 2 in all, peaking in Feb 2026 with 1.
Appearances
Like I lived there and I didn't appreciate this context.
And it's just like so recent.
I think there's a couple examples where I've heard that it's actually been started to be used.
I think...
to paint an example of why this is so much better.
For example, when GPT-5 is taking 30 minutes to respond, it's generating one token at a time.
And this diffusion idea is essentially generate all of those tokens in the completion in one batch, which is why it could be way faster.
And I think it can be suited... The startups I'm hearing are like code startups where you have a code base and you have somebody that's effectively vibe coding and they say, make this change.
And a code diff is essentially a huge...
reply from the model but it doesn't have to have that much external context and you can get it really fast by using these diffusion models so that's what i've heard of one example is that they use these text diffusion to generate really long diffs because doing it with a auto-regressive model would take minutes and that time for like a user-facing product causes a lot of churn so like every second you lose a lot of users so i i think that's going to be this thing where it's going to
grow and have some applications but i actually thought that different types of models were going to be used for different things more sooner sooner than they have been so i kind of trade off i think that the tool use point is the one that's stopping them from being um like most general purpose because like cloud code and this hatching pta was search like
the autoregressive chain is interrupted with some external tool, and I don't know how to do that with the diffusion setup.
I think there's a cool one last point on the tool use thing.
I think that you hinted at this and we've both come at this in our own ways is that the open versus closed models use tools in very different ways where open models, people go to Hugging Face and you download the model and then the person's going to be like, oh, what tool do I want?
And
I don't know, Exa is my preferred search provider, but somebody else might care for a different search startup where you release a model that needs to be useful for multiple tools for multiple use cases, which is really hard because you're making a general reasoning engine model, which is actually what GPT-OSS is good for.
But on the closed models, you're deeply integrating the specific tool into your experience.
And I think that...
open models will struggle to replicate some of the things that I like to do with closed models, which will be like, I don't know, you can reference a mix of public and private information and something that I keep trying every three to six months.
I try like Codex on the web, which is just prompting a model to make an update to some GitHub repository that I have.
Showing 741–760 of 1,814 · page 38 of 91
← Previous
Next →