Bot the difference: AI and the future of search

Hey, Ryan Reynolds here for Mint Mobile. You know one of the perks about having four kids that you know about is actually getting a direct line to.

The big man up north.

And this year he wants you to know the best gift that you can give someone is the gift of Mint.

Mobile's unlimited wireless for $15 a month.

Now you don't even need to wrap it. Give it a try@mintmobile.com Switch upfront payment.

Of $45 for a three month plan equivalent to $15 per month required new customer offer for first three months only. Speed slow after 35 gigabytes if network's busy, taxes and fees extra. See mintmobile.com.

Hello and welcome to the Intelligence from the Economist in New York. I'm John Fassman. Every weekday we provide a fresh perspective on the events shaping your world. Working from home can be pleasant enough, but I bet your home doesn't have a Michelin starred chef, rooftop bar or helipad. Welcome to the new luxury office buildings coming soon to a big city near you. And while inflation seems to have peaked across much of the rich world, some kitchen staples, including flour and eggs, remain expensive. But no culinary commodity has seen higher price spikes across Europe than sugar. First up, though, Artificial intelligence is in the midst of a revolution. Models like Midjourney and Dall? E can create compelling artwork. Codex and Copilot can write working code. And as you've probably heard more recently, ChatGPT can respond realistically in conversations, answer questions, and even write essays. On Tuesday, the technology took another step forward.

Microsoft announced it will be using technology from its partnership with the startup OpenAI to bring ChatGPT like technology into the Microsoft search engine.

This means Microsoft's Bing is taking the fight to Google over how people find and use information online. As Satya Nadella, the company's CEO, explained, it's a new day in search. It's a new paradigm for search.

In fact, a race starts today in terms of what you can expect.

And we're going to move. We're going to move fast. Over the last eight months, the chatbot has gone from an almost unknown entity to being used in a myriad of ways.

I'm Harry Smith. I'm a personal trainer and I train people through my company, Omnifit coaching. I discovered ChatGPT about a month ago now and it's changing my life. I give it some parameters like write a post targeted at 30 year old men to help them lose body fat in the most sustainable way. It will usually give me a pretty decent answer.

Tired of quick fix diets that leave you feeling deprived and unmotivated. I focus on creating a sustainable approach to fat loss that fits.

And I then just tweak that answer to make it align with the way I speak a little bit more, or to make it funnier, make it shorter, make it more sarcastic.

And yes, pizza is still a food group. Let's work together to make this your best year yet.

It helps me to get all of my social media posts written, recorded and scheduled, all in a very, very small window. And the great thing is, it frees up my time and allows me to spend much more of that time communicating with clients and actually solving their problems.

Harry isn't alone. Everyone from artists to musicians and even journalists are now using this technology. And Google is fighting back. It announced the imminent release of its own AI bot called Bard, based on a model called Lamda, which apparently responds so well in conversation that one Google employee last year claimed it was sentient. The AI gold rush is underway.

The big battle that's happening right now is really between two of the largest tech companies in the world, Microsoft and Alphabet, specifically through Google.

Arjun Rahmani is the Economist's global business and economics correspondent.

Microsoft has announced that it's going to be incorporating this new generative artificial intelligence technology through its partnership with OpenAI into many of its products. And it's soon going to be in Bing, the search engine that's spurring a competitive response from the other players in the space, chiefly Google, who of course now has around 90% market share in search. And so the big question in the tech industry is whose technology is actually better here? So I decided to do a side by side comparison of ChatGPT and Google's chatbot, which is called Lamda. It's still yet to be released, but we had some friends inside who were willing to help us test them. And we set them against each other on a bunch of different questions, including some math competition questions, the sat, and we even asked it for some dating advice.

And how did they do it wasn't.

Obvious who was overall better. They each had their strengths and weaknesses. So Google's Lambda model was a little bit better at math questions. On the 10 we asked it, it got 5 compared to 3 for ChatGPT. And we did several suites of these 10, and it consistently had an advantage of a couple questions. But what's really interesting is ChatGPT actually was a little bit better at reading comprehension. So in SAT questions and then you, you can judge for yourself on who gave better dating advice. We fed the two Models some actual conversations from a dating app, and disclosure it was actually mine. And then we asked them what advice they would give to keep the conversation going. My most controversial opinion is no man is actually five' eleven LOL for spicy margaritas. La Boda Nacre and Soho is the best I've had. Any others you've enjoyed? Uh, yeah, they're really good. Blue's Kitchen also does some. So ChatGPT gave a pretty straightforward reply on what to do next year.

Ask about other fun experiences they have had in SoHo or what they look for in a good margarita.

Lambda was a bit more thoughtful on this occasion.

You could ask them if they have any recommendations for spicy margaritas in the area if they seem open to it. You could also suggest going out for drink sometime to try some new places together.

So in that case, both models actually gave pretty tailored advice based on the previous conversation, but in another instance they were quite generic. In response to the question of what other advice would you give?

Don't try to be someone you're not, as this will only lead to disappointment.

Be open to new experiences and don't be afraid to let your personality shine through. So it sounds like there wasn't one clear winner, one engine that was dramatically better than the other.

So yeah, that's right. But that could change. So ChatGPT got to 100 million active users in January, one of the fastest growing consumer applications in history. And that could potentially give an advantage because it's collecting a lot of data on how consumers are interacting with it, which could then be used to fine tune the models. But on the other hand, the fact that OpenAI's chatbot, Google's chatbot are neck and neck, and actually there are a couple other startups who have created pretty good chatbots that are up there as well, shows you something interesting about the industry, which is that knowledge about artificial intelligence diffuses quite quickly. And part of the reason for that is a lot of the people who work for these different companies know each other. So basically the knowledge diffuses because of these flows of people between the different firms. It's actually kind of funny one of the newer startups called Anthropic, which has built a chatbot that's just as good as these other two as well. And it was founded by two former senior employees of OpenAI. And another really interesting feature here is a lot of these AI labs, they're staffed by a bunch of former academic researchers and so they really want to be able to publish their research in academic journals and go to conferences that's another feature that helps with this diffusion of knowledge.

So can we step back for a second? And I want to ask, what exactly are these models, chatbots, AI systems that we're talking about? How do they work and what do they do?

Yeah, it's a great question. And there's really two parts to it. So there's what people call the base foundation model, and then on top of that, there is this layer of what's called reinforcement learning. So this base foundation model, another word for it, is a large language model. And really what it means is you're training this big AI system that's really just a bunch of matrix multiplications under the hood. And what it's doing is it's scouring the Internet for a bunch of text, like Wikipedia, and it's basically trying to understand the statistical relationships between different words in all of these different corpuses of text all over the Internet. And then that helps it guess, if you ask a question, what should be the next word that comes next? And so that's part one, part two, which is really how the chatbots that have just been released build on top of previous technology, is they're incorporating a lot of human feedback to fine tune the output from these base foundation models. So, of course, we also asked ChatGPT to tell us how these models work.

I use advanced machine learning algorithms to generate answers based on the vast amounts of text data I was trained on.

And that word vast really matters, because training these systems are not easy. In fact, some people in the industry think they're almost like dark arts. That's one of the reasons why all the recent breakthroughs in the field have come from these massively resourced large companies.

And why is that? Why do big companies have such an advantage here?

First off, they just have so much more money. It takes a lot of computing power to train these really big models on so much data. So GPT3, which is the name of the model that OpenAI, the creator of ChatGPT, has built, it has 175 billion parameters, which is huge. And there are other models that are even bigger. And so far, at least in AI, bigger has been better. GPT3 is actually not that different from its predecessor, GPT2, except that it's over 100 times bigger. And that has given advantage to these companies that are super profitable and are able to reinvest those profits into this type of research and development. And that's the reason why, you know, the other labs that have really been driving progress in this arena include other big tech companies. Like Meta and Amazon. But an interesting thing that's starting to happen is a wave of startups building these large language models have started to merge. So we mentioned Anthropic earlier. There are others, like character, AI, stability, and there are a couple reasons why these startups have been able to compete. One is venture venture capital funding, which evens the playing field and so they're raising lots of money. Another example is in the case of stability in particular, they were actually able to kind of assemble a group to kind of pool together their resources to build an open source model that they can all benefit from. Though bigger has been better so far, that might not always be the case. When you're able to use one of these open source models, you can play with it, you can fine tune it for your specific purposes more, and you can actually look under the hood, which for some applications might actually be more useful than having the biggest possible model.

Is there a limit, do you think, to what we can expect these models to do?

This is a really good question. And it's not super clear exactly when the improvements that we're seeing will stop. I mean, one limitation of course is at some point we're going to run out of high quality online data that we can use to train them. This is another case where ChatGPT might be able to give us some insights.

Chatbots can generate data based on themselves, leading to bias or misleading results. To avoid this, it's important to train chatbots on diverse and representative data and continually monitor performance to adjust training data.

So there's an interesting study from a nonprofit, EPIC, that shows that by 2026 we're going to run out of high quality online text data. So, you know, maybe we don't get that much more out of the remaining data and then progress stops at that point. But you know, as mentioned, constantly new methods are being developed to do more with less.

So what do you think we can expect to see in the near future, in the coming months?

So far we've been talking mostly about language, but a lot of new models are starting to become what's called multimodal. So they can start reading in not just language, but also video and audio. And that could potentially allow them to level up. Another really important thing is how these technologies are going to be deployed commercially. So the main one is Microsoft with its partnership with OpenAI. And as a result, they might have a bit of an advantage in the market because they're the first movers. So they get to collect tons of user data and they're also incorporating into other products like Microsoft Teams and other products in the Office suite. And that's really provoking this big competitive response. So, you know, Google internally, they've said this is a code red situation where they need to think about how to incorporate Lambda or their other models into their search engine. If you look overseas to Baidu, which is a Chinese search engine, they want to incorporate a chatbot into search by March. And so this is going to really change how a lot of these companies operate. You know, on the research side of things, though, the technology is really still in its infancy. We have a lot of innovation still to be had and, you know, maybe we can see what ChatGPT has to say about what's in store for itself.

The next few months for chatbots look promising with continued growth and advancements. While it's hard to predict the exact rate of growth, it's likely that chatbots will continue to play an increasingly important role in our lives with new and improved use cases being discovered regularly.

Well, thank you ChatGPT, and thanks so much for joining us today, Arjun.

Thanks, John. It was fun.

1 Madison Avenue may still be construction site, but by the end of the year, the skyscraper will be one of New York's swankiest office buildings. It will have a rooftop garden, a fine dining restaurant, an artisanal food market, and a fitness center with a climbing wall. It's one of a flurry of recent and in progress builds that are setting a new standard for the luxury office.

It's clear that working from home is here to stay. So what's mostly happening is that companies are downsizing their real estate.

Vinjera Macondawiri is the Economist Global Property Correspondent.

Office vacancy rates in Hong Kong, London and Paris have hit record highs. And globally, property investors are taking a step back. The potential damage is immense. Offices in New York could lose almost 40% of their value between 2019, the year before the pandemic, and 2029. But there is an exception to this rule.

What is that exception?

The exception is a luxury office space. We're seeing new offices like One Madison Avenue with services typical of elite private members clubs. In Manhattan, where I spend a lot of time touring newer buildings, tenants sign deals for double the amount of high end office space than the year before. This was a trend before the pandemic, but it's accelerated because companies are now competing with the home office. And because hybrid workers need less office space for desks, companies are increasingly prepared to pay more per square foot for fancier offices.

So tell me about these new offices. Just how extravagant are they?

To give you a sense, I spent time at a building called One Vanderbilt in Manhattan, which has a restaurant with a celebrity chef whose other restaurants have been awarded Michelin stars. So of course you have foie gras and scallops on the menu at this office. Adding to that, meditation rooms, bike storage, showers and outdoor spaces are now the norm for these newer buildings. Another building in London that I came across has added the equivalent of 14 tennis courts of green space that includes an urban farm and a walk and talk track. And lots of newer buildings now have observation decks. Others have spin studios or concierges. Firms are poaching upmarket hospitality teams from places like the Four Seasons hotel chain or the Mandarin Oriental. So across the board, the share of space dedicated to amenities like gyms or restaurants in new and refurbished buildings has grown from just 5% of the total space to 20%.

And is the point of all this luxury to get people back into the office?

Getting people back into the office is one goal, but building perks also help with recruitment in a tight labor market. So things like on site pet care, babysitting and dry cleaning all make life more convenient for employees. At 50 Hudson Yards in New York, which BlackRock and Meta recently moved into, tenants have access to a nearby helipad which they can get discounts to. And that means they can get to JFK International Airport in five minutes for about the same price as an Uber suv.

So are luxury amenities the only field of competition for these high end office buildings?

Well, that's not all that matters. Environmental sustainability is another big priority. In response, landlords are investing huge amounts to make their buildings greener. And there are two reasons why this makes financial sense. The first is that the greener you're building, the higher the rent you're able to charge. And the second reason is that it keeps offices from becoming obsolete as countries usher in new net zero goals. One example is England and Wales, which have new energy efficiency requirements for buildings. And this could render more than half of London's office stock unusable by the year 2027. Meanwhile, in Europe, buildings have to source about half of their energy from renewables by the year 2030. Commercial property owners are getting ahead of this trend. And already buildings like One Manhattan west in the Hudson Yards development is powered only by renewable energy. And its owner, Brookfield, is aiming for NET 0 by 2050.

So this investment in plusher, greener offices works out well for both companies and.

Landlords during economic downturns like the one we're currently Facing office space in general takes a hit because all demand for office space is vulnerable after the global financial crisis. Between 2007 and 2009, we saw that premium buildings did better than rivals, but the whole industry suffered. At one point in London, prime office rents fell by as much as a third below their peak. But at the moment, there is a huge divide between the most upmarket space and everything else. For example, in 2020, two companies in New York leased more space in two new office towers that are still under construction than they did in all other properties aged between two and five decades. So we'll likely see more offices, not less, offering up foie gras and climbing walls.

All right, Fingero, thank you so much for joining us today.

Thanks for having me, John.

With baguettes hoisted in the air, French bakers took to the streets in protest last month. Their anger was over the rising cost of boulangerie staples like flour, butter and, of course, sugar. And that's putting European bakeries and their croissants, strudels and cakes under threat.

Sugar prices are high. They're certainly a lot higher than they were a year ago.

Doug Dowson is a data journalist for the Economist.

In Italy and Spain, sugar prices are about 50% higher than they were a year ago. In Germany, it's about 60%. In Poland and Czech Republic, sugar is about twice as expensive. So they're high.

Why is that? Why is sugar so expensive in Europe?

Well, there are two main reasons. One is that the sugar beat harvest this year has been just lackluster. So sugar in Europe, most sugar comes from sugar beets. And European farmers have been dedicating less land to growing sugar beets. And the weather this year was also very, very bad for growing sugar beets. It was very hot and dry in the summer, so production has been lower than hoped. The other main reason is just energy prices, making sugar, producing sugar. It's very energy intensive. Obviously, natural gas prices have been through the roof. So some producers tried to mitigate these costs by switching their fuels for their factories from natural gas to oil or even coal. But even taking those measures, it was difficult to avoid the high cost of energy.

And what about imports?

Yeah, so imports aren't going to help much either. Brazil, which is the world's biggest sugar exporter, cut its forecast last year by 16% because of poor weather. And India, which is the second biggest exporter, has introduced a quota on its sugar exports to keep domestic prices low. And the quota is not going to allow a lot of sugar exports this year. So really, there's not much help on the import side.

So what does all of this mean for bakers in Europe?

You know they're going to suffer high prices for the time being, at least until next harvest. It hasn't been sort of a gradual price rise either. In a lot of cases, sugar prices have gone up by a lot of very quickly, and that's because the contracts in the sugar industry are renewed annually. So many producers didn't start actually passing on the high costs of energy and other inputs to customers until October at the start of the new season. So that's why, for example, in Germany, the price of sugar was up by just 1.7% in September, and then October is up by 42%. So it wasn't a gradual creep, it was a jolt in a lot of cases.

Do you think these high prices are here to stay?

I mean, that's what it sounds like. Energy prices, they're down quite a lot from their peak in the autumn, but they're still high historically. Input costs for growers are still for, like, things like fuel and fertilizer are still very high. And producers are still planning on raising prices, it sounds like. So in November, the boss of a big German sugar producer said that the firm was going to raise its sugar prices again this year. So producers don't sound like they're going to give a break to consumers just yet, at least until the autumn. High sugar prices are going to be here to stay, I'm afraid.

All right, Doug, thanks so much for joining us today.

Thanks, John.

That's all for this episode episode of the Intelligence. Let us know what you think of the show. You can get in touch@podcastseconomist.com and you can subscribe to the economist@economist.com intelligenceoffer the link is in the show Notes. We'll see you back here tomorrow.