00:00We're going to talk about AI. We're going to talk about productivity. But first,
00:03we got to talk about the Fed and what we saw today. Were you just glued to the press conference
00:07sitting here on the beach? I was in our booth over there with my laptop and listening on my
00:13phone. And I don't mind the office setup today, to say the least. We got a nice breeze over here
00:18having a magical time as well. But I'm pretty pleasantly surprised with what we got from Warsh
00:25and the Fed today. I think they needed to hide. It's clear that... Fundamentally, they needed to
00:30do it. Fundamentally, they needed to do it. And I think Warsh also communicated it quite well.
00:35He was actually pretty, a bit more hawkish in his communication. But tied into really strong
00:41fundamentals. The economy is in really solid footing. And I was actually really taken aback
00:46by just how bullish he seems to sound on the economy, not only where it is today, but where
00:51it's going with AI. Do you not agree with that assessment?
00:55No, I do agree. But I think it's very inconsistent with a Fed that would want to cut rates anytime
01:00soon. So I think we've really... Like how could anybody think about cutting
01:03rates if it's... Okay.
01:04So he's laying the groundwork for the beginning of a hiking cycle.
01:09I would characterize it more as a normalization.
01:12But I think that... Why normalization rather than a hiking cycle?
01:15Because I think amongst the Fed, there's been a lot of debate around whether policy is restrictive
01:19or accommodative right now. And we got a bit of a clue from Warsh today. I don't know if
01:22he meant to or not. But he said, we are removing a dose of accommodation. And markets really
01:28perked up on that one because they were like, oh, so Warsh might actually think that policy
01:32right now is accommodative. And if that is the case, how long will it take to get to a place
01:37not only neutral, but potentially a little bit restrictive if the Fed needs to be?
01:42Stephanie, how risky do you think it is right now in terms of Fed policy that if they could
01:47they could get it wrong and push us into a slowdown?
01:50I think what they did today removes or reduces the risk of a policy error.
01:55If they didn't hike today, I think we'd be a lot more concerned about that.
01:58The reality is, as we all know, the Fed has no control of a straighter from Moose.
02:02But what they can do is begin to quell some of those inflation pressures bubbling up in
02:07the economy and also to continue to justify anchored inflation expectations.
02:12And if they just take that for granted, particularly considering that when you look at their dot
02:16plot, at the beginning of this year, they were projecting to get back to 2.1% on core PCE
02:21by
02:212028 or sorry, 2027. But now where are they? 2029, two additional years. So they have to move.
02:28How is this timelier? In terms of what did he say about things? Yeah, we want a timely return to
02:35inflation. So the reality is it's not timely at all. We are very delayed. Two years is a big delay.
02:41Yes. And they're starting to wake up to that. And so so, OK, rate normalization.
02:46You know, we think that they might have room to move 50 basis points, 75 basis points.
02:50They already moved 25. But that's not, you know, a 200 plus hiking cycle.
02:56So we don't see that in the cards. But we think that the economy strengthened.
02:59Inflation is warmer for longer. And policy needs to get back in position.
03:02How much of the inflation problems are because of higher oil prices? And that being a longer like
03:07it's not just going to go away maybe when this gets resolved, whatever that is. It's that's the
03:12new reality. I think there's this air of uncertainty just given how much of this depends on
03:17the straight of our moves, geopolitical tensions. But the way things stand right now, very little
03:22visibility as to when we're going to get an all clear, if ever. So we are looking at a stickier
03:27impact there on oil prices. And that is going to trickle through to other parts of the economy.
03:31It's not just oil, though. It's also tariffs. And it's also the AI build out. And you're seeing
03:35that with the surge in memory prices also begin to impact some consumer prices as well.
03:39But how much control does the Fed have over any of those things?
03:42Not much. But they do have control over confidence, inflation expectations, wage pressures. If
03:48consumers feel like they're in a new inflationary environment, I need to go and bang on my boss's
03:53door and ask for a raise. You know, we don't see a lot of that today. But the Fed cannot
03:56take
03:57that for granted and just sit blindly in light of this.
04:00I want to get to you in your notes. When it comes to AI, you have concerns about the
04:04AI singularity cliff. Talk to us about that. What specifically? Did I read it wrong?
04:10Maybe.
04:11Oh. I thought it was in your notes.
04:13Well, I think when it comes to AI.
04:15What is your take about AI?
04:16Maybe the bots wrote that, Carol.
04:17Yeah, maybe the bots wrote that.
04:18Maybe the bots did. They're not really good at paraphrasing, that's for sure.
04:22Okay.
04:22Which gives me some confidence that, you know, I'm still going to have a job request in time.
04:25Okay.
04:26No, look, this has been a really eventful week when it comes to AI. And I think this
04:31whole debate around whether we should slow the frontier, safety concerns, regulation,
04:35I think all of this is actually great. Because regulation is not a question of if, but when,
04:40and on what terms. And I think...
04:43Wait, so you think it's a given, the regulation?
04:45At some point, yeah.
04:47Regulation from who, though?
04:48These are the big questions.
04:49And what kind of regulation?
04:51Well, we need to have some degree of control or awareness of these safety risks. And particularly
04:57given this summer what we've seen with the model training and a lot of the unreleased
05:02models and these episodes and these, like, concerns hugging face, you know, there's clearly
05:07this growing roar, particularly in public discourse, around, you know, discomfort of where this
05:15technology is tracking. And I think what Dario Amode did is, like, read the room a little
05:20bit. Now, there's big questions as to what's going to come from this. I think...
05:24You can say something. That's a long way to action.
05:26It's true.
05:27It's true.
05:27Especially action by government.
05:28It's true. I think they probably think that it's not a question of if, but when. And
05:31if they can propose rules, maybe that will ultimately be beneficial to the lot of model
05:37labs. But in terms of what markets hear most about right now, it's really around how much
05:42they're going to spend, right? Does slowing the frontier mean that they're going to
05:45slow capex? We don't think so, necessarily. Because when you think about what capex is
05:51going towards, there's training versus inference. And this year, inference has already overtaken
05:55training. Inference makes most economical sense for these models to focus on.
06:00But is there a proven ROI for companies when it comes to the AI spend? Because I think what
06:05we started to hear is companies getting more discretionary. Maybe everybody doesn't need
06:09to have access to the models because there's a cost involved.
06:12Yeah. I think we're very early innings. I think some companies are absolutely
06:15seeing success. But the reality is, is the success of the frontier versus the success
06:21for an individual enterprise is a very, very different things. And some companies are doing
06:27it quite well. But others, whether it's because of regulation, whether it's because they don't
06:31know what services to buy, they still need to train their workers to actually use these
06:36tools. These are all really important variables to getting the ROI that have little to do with
06:41the latest model release. But the productivity question is still out there, which is, okay,
06:47well, if there is a slowing, not of the capex spend, but a slowing of this technology due to
06:53regulation, does that change the productivity outlook?
06:55I actually think slowing the frontier can help us with adoption, which is ultimately needed for
07:00productivity. Because we can pause, take a breath and understand the technology because
07:03it changes so much month to month. Exactly. It's so hard for businesses to know what to invest
07:07in right now when there's all this uncertainty of what if another model release next month comes
07:11out, completely absorbs our need to invest in that. It'll help software, it'll help hyperscalers
07:16and businesses just figure out a proper economics for this.
07:19Is it though a risk? Because China's certainly really putting a lot of pressure, the government
07:24behind their AI models and moving forward. Is there a risk of the US falling behind in terms
07:29of this technology?
07:30I think it is an overstated risk, at least in the near term. You know, I think there is a
07:35lot going
07:36for US model labs in how successful they've been in pushing the frontier, when it comes to their
07:42ability to access next generation chips. Also, the slate of just like human capital that we have here
07:49and liquid capital markets. Over the long term, yeah. But I think there is a lot that is going to
07:55be said
07:56and done geopolitically around AI standards and rulemaking that we've really yet to see.
08:01With this administration?
08:02I think regardless of who. The reality is this needs to happen.
Comments