Bulls’ AI: funding artificial intelligence

Spring is calling. Weddings, patio sessions, barbecues, and you want that drink vibe without sacrificing tomorrow. That's RK Zero Proof. As the world's first zero-proof spirits brand, RK created the warm molecule, giving you the smooth kick of whiskey or tequila with zero alcohol, zero calories, zero sugar, and all the peace of mind you need to enjoy every moment. Step into the zero-proof season at rkzeroproof.com.

The Economist.

Hello and welcome to The Intelligence from The Economist. I'm your host, Rosie Bloor. Every weekday, we provide a fresh perspective on the events shaping your world. Remember those nuclear arms agreements? That limited the number of warheads to keep us safe, well, they're actually about to expire. If you think things are grim now, they could get even grimmer. And are you tired of all the data and polls and objective measures of political fortunes? Good news, our correspondent looks at an alternative type of forecasting that's proving surprisingly popular. We meet the political astrologers. But first, these days we hear about AI all the time. It's apparently transforming every aspect of our lives. And in that revolution, one company is vastly ahead of the pack. OpenAI is the firm behind ChatGPT, and it could soon become a very rare beast indeed. Not just a Silicon Valley unicorn, but a hectocorn. That's the term for a startup valued at over $100 billion. Until now, Elon Musk's rocket company SpaceX was the only American startup to reach that level. But OpenAI may soon be valued at $150 billion. The problem is that when you get to that size and need to raise money, you really do need funders to chip in very large sums.

OpenAI's latest investment round has apparently ballooned to a new high valuation as well as a significant $250 million minimum investment.

And that means most of the usual investors just aren't big enough.

Enormous amounts of money are being poured into generative AI. And it's changing the nature of Silicon Valley.

Henry Tricks is our US tech editor.

In the last year and a half or so, venture capitalists have plowed nearly $95 billion into AI startups. But then you look at the size of OpenAI, and it's clear that this is not your usual startup. It's already become a monster. It's so big that it is changing the rules of the game in Silicon Valley.

So Henry, how is OpenAI different from previous startups?

So the first thing is that a single investment in OpenAI could cost $1 billion or more, which is way more than conventional startups. And this is because of the enormous resources that are needed to train the large language models behind OpenAI's apps such as ChatGPT. Most venture capital firms cannot afford to stump up billions of dollars in one deal. Even the biggest may do no more than about $1 billion of investment in the whole year on a variety of companies. So big tech firms like Microsoft, Amazon, Apple, they're picking up the slack. They're the ones who are providing most of the money. Microsoft has invested $13 billion in OpenAI since 2019, and Amazon has invested $4 billion in Anthropic, another AI firm, just this year. So they provide the funding, but they also provide something that is absolutely essential for the large language models, which is cloud computing. And that's why you see that in OpenAI's latest funding round, it's highly likely that Microsoft, Apple, Nvidia, sovereign wealth funds, in other words, the most deep pocketed companies on the planet are all going to take part, but not many VCs will.

Henry, does it really matter if VCs aren't taking part? Does it matter if they're being pushed out by their bigger rivals?

Well, they have alternatives. There are the companies that run on top of the large language models, all the whizzy coding companies and customer service companies which are expected to be built on the back of generative AI. So they have have a future. But the thing is, is that they feel somewhat gazumped by the big tech firms. They like to feel like they are the backbone of the entrepreneurial system in America and elsewhere. And it's with a bit of chagrin that they feel that they can't afford to play in this game. That said, there are an intrepid few who are still prepared to write really big checks, billion-dollar checks, in order to invest in OpenAI and others.

And you say they're making big bets. How much money are we talking about here? How much does it cost to set up and run these AI models?

By some estimates, the costs are spiraling exponentially. One estimate suggested that when ChatGPT was released in 2022, the models typically cost $10 million or less to train. Fast forward to now, and the equivalent of OpenAI's GPT-4 model costs anywhere from kind of $50 million to $100 million or more. And what people are doing is extrapolating from that, i.e., these models getting 10 times more expensive every generation, and saying that pretty soon, i.e., within the next year or two, it may cost $1 billion or more to train a model and perhaps eventually $10 billion or more. These are what they call scaling laws. And some people are skeptical about how predictable these scaling laws really are. But this is the basis on which people are prepared to put enormous sums into the models because they feel like as they get more expensive, they get better and better, and the better that they get, the more chance they have of kind of conquering the world.

So the assumption in what you are saying, Henry, then, is the more money you put in, the better it gets, the better AI keeps getting. Is that right?

It's a good question around which there is a huge amount of debate. For a start, there are physical constraints to building bigger and bigger models. It's pretty well known that the models need huge amounts of energy, chips, cooling systems in order to do their magic, and that may slow the process down. But the interesting thing is to see how these constraints are already stimulating bursts of creativity, which after all is what Silicon Valley is supposed to be really good at. So Developers are now tweaking the software and the algorithms that support the chips on which these models are trained in order to make them run more efficiently. There are also smaller and more specialized models that are being distilled out of these huge ones, and they're proving to be both cheaper and pretty effective. In the last couple of weeks, OpenAI released its newest model, GPT-4.1, or better nicknamed as Strawberry, which is designed to be better at reasoning, though it's very different from ChatGPT. It doesn't generate text. So, yeah, all of these innovations are making a variety of models which are each suited to different types of problems. And the cloud service providers on which these models run, they use all of these models in order to make answering a chatbot query cheaper and better.

And where does all that creativity leave OpenAI, which is so dominant at the moment?

Well, I think it's fair to say that OpenAI is definitely at an advantage, given the fact that it was a first mover and it has Microsoft underpinning it. Lots of money, lots of computing power, and lots of brainpower. You know, it does, continue to be the pioneer in the field, as we saw with the release of this new model. But that doesn't necessarily mean that it will become the model to rule them all. And I think no one really wants that. This is a dangerous world, generative AI, and you really want a lot more competition rather than a lot less competition. And we're seeing that competition play out. There are other big model makers like Anthropic, Google, Meta, and Elon Musk Google is coming up the rear with xAI, and they all have access to a lot of resources and are all trying to catch up. So if the trend continues, I think it's fair to say that the AI universe will be a constellation of models rather than just a single superstar.

Henry, great to talk to you.

Thank you, Rosie. Great to talk to you too.

For military planners at the Pentagon, the ultimate aim is to be prepared for every threat so that they don't have to confront any of them. That task is getting a lot harder. For decades, nuclear strategy has been about keeping the size of arsenals in check. Now, alliances are more complex than they were in the past, and more fluid. So for the first time since the Cold War, American strategists are questioning the current nuclear order.

We've had decades of reductions in the world's stockpile of nuclear weapons. They've come down from about 70,000 to about 12,000 thanks to successive arms control agreements. And we're now seeing a world in which that is reversing.

Anton LaGuardia is our diplomatic editor.

China is building up, and the United States, for the first time in a very long time, has said it is getting ready to increase the size of the stockpile if the president should order that.

So just explain, what exactly is the new threat?

The new threat is both in terms of quantity and in terms of complexity. First of all, the existing arms control agreements essentially limited the world's two largest nuclear powers, the United States and the Soviet Union, and now Russia, to be roughly equal with about 1,500 strategic warheads deployed. And those two words are important: strategic and deployed. Strategic means long range. Deployed means actually ready to use as opposed to held in stock or in reserve. And the problem is that China has until recently had a very small arsenal of a few hundred warheads and is now building up fast, according to the Pentagon, to 1,000 warheads by the end of the decade and perhaps 1,500 warheads by 2035. And China and Russia are much closer buddies than they used to be. China is helping Russia, short of providing actual weapons, but supplying the defense industrial base for its war in Ukraine. Moreover, other nuclear powers are also more closely bound with Russia and China, North Korea especially, but also Iran, which is a nuclear threshold state. So altogether, the American worry is what happens if all these rivals gang up on us at the same time, or what happens if there is a chain reaction of conflicts. So one Pentagon official described to me what he thought was his nightmare scenario: a war in Europe which involves NATO. Russia attacks NATO, perhaps the Baltic states, Draws in the Americans. They get involved in Europe. China takes the opportunity to invade Taiwan, thinking the Americans are now stretched. And the North Koreans seize the opportunity to reunify South Korea because they see the Americans even more stretched. So could the Americans deal with all those crises? And indeed, could they deter them beforehand? And the Americans are just not clear anymore that they can do so.

You've laid out such a terrifying threat of nuclear Armageddon, I'm not quite sure where to go from there. But what's happened to the treaties that were supposed to keep the stockpile of nuclear weapons relatively contained?

I hate to frighten people even more, but the worrying thing is that the main treaty that limits the weapons of the US and Russia, the New START treaty, is about to expire in less than 2 years' time. In February 2026, and there is no prospect at the moment of a follow-on agreement. Essentially, nuclear talks between the two sides have broken down, and the Russians and the Americans have more or less stopped abiding by the treaty's verification procedures, although they say they still abide by the numerical limits that those treaties impose. In other words, The deployed number of warheads is still the same according to each side, but it's more and more difficult to verify that that is taking place because there are no more inspections. They don't report their numbers to each other. They don't report the movements of nuclear weapons and nuclear-capable missiles and so on and so forth. So as time goes by, there's much greater uncertainty, and obviously that creates space for mistakes and miscalculations.

So how is the US proposing to resolve this?

On two tracks, really. One is to say they continue to be interested in arms control and risk reduction, but also to hedge for the future and to get ready for when the New START treaty expires in 2026, if the president decides to do so, that they are able to build up quickly the size of the deployed stockpile. That means in the first instance, taking things out of reserve and putting them on intercontinental ballistic missiles or putting them onto submarines or onto bombers. And that could potentially give you a very quick increase, potentially doubling the deployed stockpile. If you wanted to increase the overall stockpile, the total stockpile, then that means building many more nuclear weapons, which, you know, involves more effort and more time, not impossible, but would take more time. And the Americans are currently engaged in a complete modernization of their whole existing nuclear weapons enterprise. So they've got new bombers being built, they're building nuclear ballistic missile submarines, and they're building new ICBMs. A lot of these programs are overcost and delayed, and they're also building a whole new command and control system. That's all allowed within the current arms control systems, but they're modernized, more capable, and so on. So there's a lot of money already going in, and the further step is to say, how do we reverse the engineering that we did years ago to reduce the numbers that are deployed?

Are we heading into a second Cold War, Anton?

In many ways, yes. It'll be different in that China is much more economically integrated with the United States and the West, so it's a different kind of competition than the West had with the Soviet bloc. But I think on this question of nuclear weapons, we're in a slow-moving arms race that risks accelerating from 2026. There is always the worry that as one side builds up, the other side feels it needs to build up in response. There are people that will argue that the United States has more than enough nuclear weapons to deter both China and Russia and others does not need to hit every nuclear weapon silo. It just needs to hold enemy cities at risk. And there are people in the arms control community who will say, look, America's thousands of warheads do not buy you that much more deterrence than Britain's or France's couple of hundred warheads. Each. I pray and hope it doesn't reach the astronomical numbers that they reached during the Cold War with the ability to destroy the world several times over. And I hope that the lessons that the superpowers learned in the '60s are remembered and reapplied to the present day. But I'm afraid that for as long as this rivalry with China continues to grow and deepen, and Russia's war with Ukraine remains unresolved, and continues to go poorly for Russia, we are in an age of higher nuclear risks all around.

Anton, thank you. It's fascinating and terrifying to talk to you.

Thank you, Rosie.

The thing about the election is that so many planets are retrograde, um, right after it. Like Mars, we're gonna have a Mercury retrograde, Jupiter's retrograde. Like, it's just like so much going on. What does that mean? What I think it means is mass confusion.

That's Helena Woods, a professional astrologer.

Holly Berman is a news editor at The Economist.

I recently spoke to her to find out more about how astrology works and how it can be used to predict all kinds of things, including elections.

So how I'm doing this is in the middle ring, The first kind of ring here, this is a chart, and I can put someone's birth chart here. So like, as an example, this is Trump's chart.

Helena showed me Donald Trump's horoscope on a shared screen, and what I could see was a wheel of zodiac symbols, lines, and colors, all of which she said was a snapshot of his destiny determined by the positions of planets across different aspects or houses of his life. —so that includes things like relationships, his career, and by default, the upcoming election. To get the most accurate reading, Helena needed Trump's exact time of birth, a detail that she finds online.

Astro.com, they have all, like, any political figure, public person, they have all of the information. Trump, he's got major things happening. He's going to enter the peak of his life on April 25th.

Apparently, Kamala Harris has pretty big things happening to her too. Jupiter and Saturn are aligning in her favor.

So if she does win, I do think she might carry out two terms.

Astrologers have long claimed to be able to chart the fate of politicians. One advised Ronald and Nancy Reagan ahead of meetings with world leaders. But now astrologers are captivating a new generation. In 2022, 37% of Americans aged between 18 to 29 surveyed by YouGov said they believe in astrology. Helena is one of many astrologers using social media to reach them. Hey there, welcome back to the channel Astrocartography with Helena, where we talk all about the astrology of place. She has nearly 100,000 subscribers across two YouTube channels.

Kamala Harris, does she have a chance of winning the presidential election based on her astrology?

Let's get into it. Election cycles are an opportunity to go viral. Evan Nathaniel Grimm accumulated 400,000 TikTok followers and millions of views filming horoscope videos and cosmically informed forecasts about the news.

If you have your sun, moon, or rising in Cancer, Leo, Virgo, Libra, Scorpio, or Sagittarius, then this is what's coming for you.

Evan told me he had a long list of accurate predictions, including that a very volatile event was on the cards for Donald Trump before the assassination attempt happened in July.

I did talk about how Trump would go through a very volatile event July 15th, and July 13th was the actual assassination attempt. I predicted that Joe Biden would drop out on July 23rd. He dropped out on the 21st. And, you know, I predicted the end of the Hollywood writers' strike last year. I said it would happen.

Evan does not see Trump on the ticket after October, owing in part to what he says is an upcoming Aries full moon. Earning a good reputation can be lucrative. Evan told me he's completed thousands of private readings for paying clients.

My all-in-one readings, which combine the birth chart and the transits and the secondary progressions, which is another timing technique, that is $750 for an hour.

Many astrologers put their juiciest forecasts behind paywalls, and the value of the global astrology market is expected to reach $22.8 billion by 2031, up from $12.8 billion in 2022. But even astrology is not without polarization. Helena told me that she has paused making political videos because they were causing intense division on her channel. Another astrologer told me she's been cancelled over some of her interpretations. And when I spoke to Helena, she also told me that astrologers can predict the future of a country. Turning to America's birth chart, she made some various predictions for the coming years.

Um, so this is kind of when things get wild, and then things get really insane once we get into July of '25.

In what way?

What do you, what do you see there? Um, mass division amongst people, um, fighting, conflict, independence, wanting to break free.

When I asked her when all of this might calm down, she told me the year 2033. Hang tight until then.

This year, more than half the world's population has already or will cast their votes in over 70 elections worldwide. Tomorrow, our editor-in-chief, Sanni Minton Beddoes, will be live in New York to discuss these elections and their impact on democracy. Economist subscribers can click the link in the show notes to sign up. That's all for this episode of The Intelligence. Let us know what you think of the show. You can get in touch at podcast@theintelligence.co.uk. Economist.com. We'll see you back here tomorrow.