Because Los Alamos National Laboratory required extremely high security and clearance for national security research, OpenAI could not deploy its models through standard APIs and instead built a custom on-prem deployment, physically installing the o3 model's weights onto the lab's air-gapped Venado supercomputer.

14:24very interesting about this one was that it's  also a story of a very, like, bespoke and, like, new type of deployment that we've done. So because  they are so, they're a government lab, they're so, you know, restrictive and high security and high  clearance with a lot of their things, we couldn't just do a normal deployment with them. They  couldn't, you know, you can't have people doing national security research just hitting our APIs.  And so we actually did a custom on-prem deployment

14:48with them onto one of their supercomputers called  Venado. And so this actually involves a bunch of, you know, very bespoke work with some FDEs,  also with a lot of our developer team, to actually bring one of our reasoning models,  o3, into their laboratory, into an air-gapped, you know, supercomputer Venado and actually deploy  it and get it installed to work on their hardware, on their networking stack, and actually run it  in this particular environment. And so it was

15:16actually very interesting because we literally  had to bring the weights of the model physically into their supercomputer in an environment,  by the way, where you're not allowed to have, you know, it's very locked down for a good reason.  They're not allowed to have cell phones or like any electronics with you as well. So I think that  was a very unique challenge. And then the other interesting thing about this deployment is just  how it's being used, right? So the interesting

15:39thing is because it's so locked down and on-prem,  we actually do not have much visibility into exactly what they're doing with it, but we do  have, you know, they give us feedback. Yeah, yeah. They actually do have some telemetry, but  it's, you know, within their own systems. But we do know that it's, you know, being used for a  bunch of different things is being used for aiding

15:59them in terms of speeding up their experiments.  They have a lot of data analysis use cases, a lot of notebooks that they're running with  reams of data that they're trying to process. They're actually using it as a thought partner,  which is something that's pretty interesting to me. o3 is like pretty smart as a model. And a lot  of these people are tackling really tough, you know, novel research problems. And a lot of times  they're kind of using o3 and going back and forth

16:20with it on their experiment design on like what  they actually should be using it for, which is, you know, something that we couldn't really say  about our older models. And so, yeah, it's just being used for a lot of different use cases for  the National Lab. And the other cool thing is it's actually being shared between Los Alamos and some  of the other labs, Lawrence Livermore, Sandia as