Upcoming Rate Cuts Will Send Stocks, Assets MUCH HIGHER From Here

0:00Hello, everyone. Stocks are booming, jobs are disappearing, the president wants interest rate cuts, and AI is inventing new jobs that's going to pay you more money than your current job. We're live today from the desk of Anthony Pompliano.

0:22Before we get into today's episode, I need your help. We currently have 44 ,155 subscribers on this channel, but with your help, we'll add one. Hit the button and let's get into today's episode. All right, ladies and gentlemen, the investment community is very bullish on stocks right now. Yet at the same time, they believe that economic pain is ahead. According to a recent Gallup poll, approximately 50 % is one out of every two Americans. They expect that the stock market is going to rise over the next six months. Adam Kobese points out that this is the second highest reading since 2020. You can see the excitement in the trading volume data too.

0:57Walter Bloomberg explains, U.S. equity markets are seeing record turnover. It's averaging$1 trillion a day in the month of January. It's a 50 % jump from a year ago. And they've got 19 billion shares that are traded daily. That surge spans retail, institutional, and automated trading. It's occurring amid calm markets rather than highly volatile markets. And the factors include higher stock prices, sector rotation from tech into energy and materials, the growth in ETFs, and retail platforms like Robinhood. But in contrast to this enthusiasm in the equity market, 50 % of Americans, one out of every two, also expect unemployment to increase.

1:35That's the highest percentage since May of 2009 in the global financial crisis. It's even higher than the 49 % seen during the 2020 pandemic. So we've got a widespread belief that stocks are going to keep going up, even though companies will be laying off a meaningful percentage of their employees. The only good explanation for this paradox is the rise of artificial intelligence. Public companies are openly telling their shareholders, We will create more revenue and profit, yet we're going to need less employees to pull it off. Private companies are hiring fewer open roles. Software and many other roles, they're just falling off a cliff.

2:09And the anecdotal reports from the front lines of entry-level employees looking for a job, that paints a pretty tough picture. The data is only going to get worse too. Anna Wong, the chief U.S. economist at Bloomberg, she writes that we are expecting about a$666 ,000 downward revision to the March 2025 payroll level. It's all part of the annual benchmarking. After all that's done, we also expect December 2025 to see a downward revision of nearly 1 million jobs. It's crazy. So stocks are booming and jobs are disappearing. What should the Fed do? Well, they are going to have to cut interest rates and they can't risk the pain from deflationary forces forcing the United States into a deflationary spiral.

2:49We've got tariffs, AI and robotics swallowing the U.S. economy at an accelerated pace. Experts in the AI field are literally writing pieces comparing the pervasiveness of AI to the unexpected spread of COVID in 2020. So let me be very clear. Everyone is underestimating what is happening right now. This is not some simple change. We're watching two major changes happen to the U.S. economy at the same time. The private sector is automating significant amounts of the economy at a pace we have never seen in human history. And the public sector is re-engineering the global monetary order with tariffs, deregulation, tax cuts, and a weaker dollar.

3:25You have the President of the United States literally calling for a 200 basis point cut to interest rates so that the government can erase the annual deficit. Just watch this video here. Interest rates should be lower. You know, every point is$600 billion. Think of that,$600 billion. All he has to do, if we went down two points, we don't have a deficit anymore. And that's without cutting. and it's just a paper charge. When you think about it, it's a paper charge. We should be the lowest interest rate in the world. Now, if Trump, Besson, and Kevin Wersh are all advocating for large interest rate cuts, what do you think is gonna happen?

4:03Do you think that they're gonna cut rates? Yes, I think so. They're going to implement their view of the world. But this brings me back to the unlikely situation of stocks going higher or the job market evaporates. Usually, GDP surges and stocks lag. Then central banks raise interest rates to cool off the economy. But we have GDP booming, stocks pushing higher, and interest rates are being cut. It's going to only mean one thing. Make sure that you are holding assets instead of cash. They're going to weaken the US dollar like we have never seen before. And the short-term deflationary pressures, it's going to hide that devaluation very, very well.

4:39So Bitcoin and gold crowd, you're going to be very happy in the long run. But there may be some headwinds in the short term. And that's the true test of conviction. Can you ignore all of the noise in the short term, but keep holding? I think that those people who can do that, they're going to be the ones that are happiest a decade from now. Speaking of artificial intelligence impact on the job market, we've got a very special treat for you. We have Micro One CEO and founder Ali Ansari. He's going to sit down with us today and explain a brand new sector that did not exist just a couple of years ago.

5:10This is a very fast growing company that is doing$200 million in annual recurring revenue. They are growing at an insane pace and they are pioneering a sector where people will say, if AI becomes pervasive, this is where new jobs are going to come from. Here's my conversation with Ali Ansari. All right, Ali, a great place to start the conversation. These model labs are obviously trying their best to train data, but you've got a unique view as to how they can get data to train with. Explain what the existing model is and then how you guys are doing something a little bit different. Yeah, so the existing model has been you take very large data sets and you train on them in kind of an unstructured manner and you get to a good place.

5:50But really the way that you expand on that and make models useful across many economically valuable tasks is by getting experts that do those tasks in their day jobs and are really good at doing those tasks and having them kind of create what's called structured human data, which is basically structured human judgment from those experts. and then distilling that into the models. So that's essentially what we do is we kind of, you know, we source and vet these highly skilled professionals that are PhDs, professors, doctors, lawyers, et cetera. And we help set up data pipelines to train LLMs. And when you look at this, what is the problem with using the unstructured data, right?

6:29Like obviously some models may still use both kind of approaches, but what are like the pros and cons of each approach? Yeah, so the problem is large unstructured data sets that exist are simply not high quality. Like if you when you train on the Internet, there's some level of quality that exists and any equivalent large data sets that is in the similar order magnitude of the Internet as a whole is not going to have a materially different quality. So the only way that you can get models to be really good at medical finance or whatever domain you're trying to turn them on is to incentivize experts to specifically create data sets that are very specific taxonomies that researchers have come up with that will improve the models.

7:16And the quality of data difference there is what actually gets the models to improve. And talk a little bit about this like kind of large industry. I mean, it's much larger than I think I thought it was of how many people are working as, you know, essentially like model trainers. Yeah, exactly. AI trainers is the term. There's a lot, you know, there's over 100 ,000 people around the world and actually a good portion are in the US that are working, you know, across multiple different vendors helping labs train models. and a good portion of that is sort of like generalists, but now the largest portion is actually experts across like 100 plus domains.

7:57And what we believe is that this is going to be one of the largest job sectors that actually gets created because of AI and it's already a pretty large job sector, but ultimately there's going to be many millions and maybe even tens of millions of people that focus a good amount over their week on creating this kind of structured data for labs. And then one example that we kind of like to use that conveys the picture well is right now at MicroOne, we have about a thousand lawyers that are working, creating human data for our customers. And they're getting paid roughly 20 % more than what they make at their law firm.

8:34And so the difference is they're creating unstructured data for their law firm, which is the customers at their law firm. And they're creating structured data for MicroOne for the labs, which help them automate parts of the work that they do. and the you know the cool part is that the economy has sort of decided that that lawyers time is actually better spent more well spent uh creating structured data which is why they're paid higher and of course this doesn't mean that every lawyer should only spend time on structured data because the economy must continue functioning obviously so there's this kind of equilibrium point which is some percentage of lawyers can spend time on this and that will continue to go up and lawyers just want to explain maybe a little bit more detail you take this lawyer example when you talk about unstructured versus structured data.

9:16Like what, what do you mean by structured data? What does that look like? Yeah. So imagine a lawyer is, uh, you know, redlining a contract, uh, and sending it back to their customer. Um, they, they do so on, you know, word and they kind of email it back to the customer and the customer gets value out of that. And then that document is no longer really used in any way. Um, now that's unstructured data creation, which is obviously valuable. But what is arguably more valuable is a lawyer in that same context, instead of creating the redlining for that one document, the lawyer would create a prompt, which is, you know, defines the task, which is to redline that document.

9:57And then instead of just doing the actual redlines, they would do the redlines, but they would also create what's called rubrics that define exactly what the perfect redlines would be. So the rubrics would be a bunch of items that say, okay, the red line must include this, the red line must end with this, et cetera, et cetera. And that right there, that kind of rubric items becomes a structured data point for models to train on and help that lawyer, next time they redline a similar document, help them do it faster. Now, when you take the kind of prompt and the rubric, where does it go from there?

10:32So that a single lawyer creates that for just red lines, you then pass that to a model lab or do you need a certain number of them? Like walk me through kind of the process of you've now paid a lawyer to do this. You've got the prompt in the red line. What happens next? Yeah. So usually these data sets are what's called evaluation data sets. So what happens is each rubric is essentially defining how many points the model response should get based on whether or not they actually meet all the rubric items and so if they you know if the model response red lines very well and they meet 70 of the rubric items they get a 70 score or if they actually don't do the red line well and they only meet one of the criteria they get a very low score so the what happens is you know this example of lawyers there will be hundreds of lawyers that create lots of these lots of these different tasks which are prompts and and rubric outputs.

11:28And then what we would do is then run a large number of these prompts through models and see how they actually perform on them and what scores they get. And those scores become evaluation kind of benchmarks. And then those benchmarks become training data for the models. So I think the best way to think about this is all of the benchmarks and evals that happen publicly and privately and so forth, those are all training data sets for models to actually improve because those benchmarks result in reward models. And then the policy models of these companies connect to those reward models to actually improve.

12:03And when you're doing this, how do you ensure accuracy? Is it just on the front end of making sure you get the smartest lawyers? Is there some sort of quality check as it's going? Do you build a really large data set of a bunch of lawyers doing this all at one time and kind of benchmark them? Yeah. So quality and accuracy and kind of like striving for the objective truth in any domain is really the main is really the main kind of infrastructure that we've built. And there's a lot of aspects to it. One of them is we have peer reviews. So similar to how there's like peer reviewers in the research process, there's peer reviews to data creation.

12:39So we have, you know, one lawyer creates some, some response, some rubrics or whatever the data structure is. There's then another lawyer that kind of checks that. And then sometimes there's kind of a tiebreaker, ultimate peer review. So that's one thing. The second thing is we, we try to simulate real world tasks and real world environments as much as we can because the the way you if you're able to simulate like what the actual workflow of lawyers working at a law firm uh looks like then that gets closer to what is ultimately the most accurate uh you know outcome which you know customers would would essentially pay for um so peer reviews plus kind of simulating real worldness as much as we can is is what gets these to be uh you know less subjective and and more truthful.

13:24Got it. And then talk about the economics of this thing. So you mentioned that paying the lawyer 20 % more than they could earn at their job. How do you guys make money? Why are the model labs paying for this? Couldn't they just go find the lawyers themselves? AI labs have, they have vendors for this and they're not building this in-house because there's two main reasons. One is the technology required here is three pillars. One is AI recruitment engine which basically sources of vets experts two is a data platform to actually like create the data review it and so forth and then ultimately there's a talent performance management product that kind of quantifies the expert data quality velocity and all other metrics and to to build these in-house is I mean it's a difficult technology to build but also it's a distraction from the core model build outs so that's the first thing the The second thing is to get the amount of data required here, you need in most pipelines, you need hundreds of experts and sometimes you need thousands.

14:26And so if labs were to do this in-house, they would have to hire many thousands of people, you know, full-time in-house that are, that end up kind of replacing the core composition of their team from researchers to domain experts across like a bunch of random domains. So they are no longer a research lab if they take that route, which is why labs essentially outsource to companies like MicroOne. And the payment model is essentially, there's some amount of data points that are required month over month. And it's basically a data consumption model labs pay for per data point. And then we kind of pay the experts on our end.

15:08I see. And then as you're going through this, it feels like really what they're doing is they're doing their job. Plus, they are then doing extra work, right? They're creating the rubric, all this kind of stuff. Is this a thing in maybe other industries where most people are going to end up just training the machines? Like a lawyer still feels like a job where there's some advice, you know, and maybe some decision making, some taste, right? Some intuition that's important. I don't know if we're yet ready for robots to go into court and argue with each other yet, but maybe one day. But you could easily see other industries where, you know, you may just be able to automate almost the entire industry.

15:45And so how do you think about the labor market impact of something like this? Yeah, I think ultimately a very material percentage of the entire human labor market will go towards structured human data creation. and i think you know the the way that we think about it is um the ultimate goal for these models is to build agents that can automate functions within the economy and ultimately the goal would be to automate all functions within the economy of course first there's kind of digitally intelligent models and then ultimately there will be models that can act in the real world which uh which would automate kind of real world functions um so i think there is a there is an ultimate strive to you know fully automate the the entire economy as it currently exists but but i think the nuance is um as we iteratively automate these functions the humans that work in those functions are you know their time is going to get freed up and they they will come up with kind of new things to work on that are higher level that are more creative that are more fun for humanity and they will essentially come up with net new functions and and then those net new functions is what humans will spend, you know, kind of creating unstructured data for, which is like, in other words, like they're just doing their work.

17:02And then we'll continue this loop of like those net new functions, then being useful to be automated, which then structured data needs to be created. And, you know, this sort of continuous loop will keep going. And so I think the, at some point we'll reach a kind of an equilibrium where some percentage, you know, maybe 5%, maybe 10 % of all human labor gets spent on structured data creation, while the rest is spent on kind of functioning the net new functions that are not yet automated. And, you know, we sort of think that there's actually no last mile here. There's no last mile of automations.

17:37It's more so just this like, you know, continuous infinite strive towards automating everything new that is coming up. I mean, 5 or 10 % of the human labor force is a big number. That's a lot of people helping to create this structured data? Yeah, it is a lot. I think eventually we will have tens of millions of people working in some way in AI training. And I think some folks will actually, in fact, do a full-time. And even right now, there are some folks that do a full-time because you're not only making your job more fun as models actually get good at helping you with your job but but you're also really doing what you love like you're essentially able to do the most pure version of your work when you're creating these these data sets because there's no like operational headache of uh doing mna when you're creating rubric for mna you're just focused on the the pure art of that redlining for that mna and of course this is not just for like you know lawyers it's across every domain when when when an artist is like you know giving their preferences on which art is better or they're creating some sort of image that will help the models train.

18:54There's no kind of operations outside of the pure art of doing that task. So I think there's a beauty to allowing humans to actually do only the thing they like, which is the pure functions of any given domain. It makes sense to me. What about your company? How big is it? What numbers have you shared publicly? Yeah. So we're about 85 people. We've surpassed 200 million run rates and we're growing. The growth is kind of accelerating more than 30 % month over month. And yeah, so it's been a fun last few quarters. I am very fortunate to be a very small investor in your company, but I need you to back up for a second.

19:36You started the company four years ago, you're doing over$200 million of annual revenue and you are growing 30 % month over month? That's right. We actually, we publicly mentioned the 200 million run rates last week, and now it's quite beyond that. And so the growth is accelerating. And I think we might hit for the next few months, we might even hit above 35 % month over month growth. so i mean you're adding you know a million bucks a day or more of arr that's pretty impressive yeah it's a few million bucks a day all right where uh where can we send people and maybe uh help explain like who are the ideal customers right uh on the enterprise side but then also i'm assuming you guys are looking for people so what are the industries who are the types of people do they got to be smart do they got to live in a certain place they got to have a certain type of computer like you know who are looking for yeah so we're looking on the expert side we work with uh doctors lawyers and finance experts the most we have a very good portion of our current active pipelines in those domains so doctors lawyers and finance experts all around the world and lots of different subdomains you know feel free to kind of sign up on micro one but also on the on on the core team side we're hiring a lot of researchers and engineers i mean that's uh you know kind of the main teams that were scaling up very quickly.

21:03So AI researchers to work out of our SF office, as well as full stack engineers to kind of help build these platforms are kind of the main hires. Got it. And then what about potential customers? Yeah, customers is, we have three customer types. One is, of course, the labs, which we're grateful to pretty much work with all the frontier labs. The second is for our data vertical that we call Cortex, which is essentially contextual evaluations for enterprise agents. um that is kind of similar similar to what we do for the labs but instead is to like implement the agents within the enterprises so you know the fortune 1000 broadly would be customers here so um yeah that would be second and the third is uh robotics which is like real world robotics data for robotics labs makes uh makes sense to me all right man you uh you're killing it it's uh it's very cool to see and uh you know it's a classic story of not only did you pick kind of the right sector and to go after but you guys all obviously have executed very very well so So congratulations on all the success.

22:02And we'll definitely do this again in the future. We'll do it again when you're at a billion in ARR. You can't come back until you hit a billion. So we'll be back into this year. I wasn't going to say that, but you did. All right. See you soon. Thank you, Anthony. Appreciate it. Now, I don't know about you guys, but hearing Ali talk about this thing, it's pretty cool to see that artificial intelligence may put some people out of a job, but it's also creating net new jobs as well in sectors where there are professional white-collar workers. And so hearing him talk about this gives us a little peek around the corner into the future.

22:36That's it for today's show. Thank you guys so much for watching. Please remember to subscribe on YouTube, and I'll see you guys live tomorrow from the desk of Anthony Pompliano.