Meta’s Muse Wonder Challenges OpenAI, Why Companies Struggle to Prove AI Returns

Akash Pasricha (0:13)

Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Tuesday, October 13th. Today on the show, we are going to unpack how the White House is considering taking more ownership over the task of reviewing frontier AI models. We're then going to look at Meta's new Wonder models and hear about how they're performing. We're going to dig into a new study from UPenn, showing businesses are struggling to measure the returns of AI.

And to close out the show, we have got new data showing that developers are spending less on leading edge AI models and are choosing cheaper models that are just good enough. It's going to be a great show, so let's get right on into it.

The Center for Advancing Innovation and Standards for Superintelligence, that is a real team within the government and a very important one to pay attention to, because they are the ones that are reviewing AI models before they come out for security risks. My colleague, Leo Schwartz, published a scoop that the Trump Administration is thinking about moving part of that group into the White House. I want to bring on Leo to break down his reporting. Leo, welcome back to the show. It's great to have you here.

Leo Schwartz (1:27)

Thanks for having me.

Akash Pasricha (1:28)

Okay. So, Casey is the acronym for the group that I'm not going to say the name of again, because I will never remember what it is.

Just remind us, so where do they sit currently within the government?

Leo Schwartz (1:43)

Yeah. Well, you know I wouldn't come here without a new acronym, so I'm glad I have one. It's actually pronounced the same. It was Casey before. It's now Casey, but of course, you have AI switched for super intelligence, which is what Trump is calling it.

Casey is within the Commerce Department. It's long been the subject of political debate and turmoil, including for months, really, since the beginning of the Trump Administration, debates over whether it should be moved out of the Commerce Department where it sits under yet another acronym, NIST, or the National Institute for Standards and Technology. There's been discussions to move it to the Department of Energy. Then in the past months about moving into the White House, where it would be less frozen out of conversations and more integrated into the White House's core policy-making decision-making around AI.

Akash Pasricha (2:34)

Why did it initially sit within the Commerce Department? What was the thinking there?

Leo Schwartz (2:40)

It was created under the Biden administration as its original name, the AI Safety Institute. Of course, when David Sachs came in, former AIs are, they didn't like all this talk about safety, so they changed its name. But it made sense to be under Commerce at the time because of that center that houses it, the National Institute for Standards Technology that does a lot of the science testing within the broader executive branch.

There was an idea that this institute, which was renamed to Casey under Trump, would be really important for evaluating, testing, coming up with standards for AI models. That, of course, only grew in importance over the last year, and you saw reorientation within the Trump Administration, where they realized that Casey actually was incredibly important. It's where some of the administration's top AI talent sat, but it's long been starved of resources and not part of that core AI decision-making.

Akash Pasricha (3:38)

I want to get to the White House decision here in a second, but one thing you did mention in the story was that they were considering moving it to the Department of Energy, which you just mentioned here. So what was the context in the politics around that?

Leo Schwartz (3:53)

So we reported in this story in a while back that Michael Kratzios, who's the Director of the Big Science Technology Policy Office within the White House had previously advocated for being in the DOE, mostly because the DOE has access to all sorts of resources, like for example, it's National Laboratories where it would have better access for staff and hardware resources to be able to do its testing. That didn't happen. It does seem like there's always been a bit of political maneuvering at play, where obviously if you're a cabinet official or you're overseeing any agency, you don't necessarily just want to lose it to a different office or agency.

Akash Pasricha (4:33)

Right. Okay. So now the White House comes into the picture.

They want Casey all for themselves. They want part of Casey. What did you find?

Leo Schwartz (4:43)

So I think the big thing is basically when within the White House, most of the staff is actually housed in a couple of buildings. If there's decision-making, if there's meetings happening about how to do model testing, how to approach supervision of AI models, what's happening with this new SI task force, it would actually make sense for at least the leadership of Casey to be in the vicinity, to be at the White House, to be reporting directly to these key stakeholders, namely Michael Kratzios from the Office of Science and Technology Policy, also Sean Carencross, who leads the Office of the National Cyber Director, which is overseeing a lot of AI policy-making and supervision within the White House right now. It would make sense for Casey, who has that highly technical AI staff, to at least be reporting directly to the White House rather than being a little bit more frozen out or a bit more auxiliary at the Commerce Department.