The global banking system is experiencing a profound transformation. Traditional banking infrastructure is rapidly losing ground to the ubiquity of mobile phones and the rise of agent-based and telco-led financial services. The numbers reveal a structural shift that is bypassing banks—especially in rural and low-income markets—at an unprecedented scale.

ATMs vs. Mobile Phones: The Scale of Displacement

There are just 2.95 million ATMs worldwide, compared to 8.3 billion active mobile phones (BankMyCell, 2025). This means that for every ATM, there are about 2,820 cellphones. In rural areas, which account for 43% of the world’s population (World Bank, SP.RUR.TOTL.ZS), the ATM is increasingly irrelevant. The future of financial transactions is mobile, and the numbers confirm it.

Deposit Outflows and Digital Growth

In 2024, U.S. banks recorded their first deposit decline since 1995—a 4.8% drop (FinTech Weekly). Meanwhile, digital-first banks such as N26 reported revenue growth of 40% in the same period. Over 55% of U.S. customers now primarily manage their finances through mobile apps, and the global digital banking market is projected to reach $22.3 trillion by 2026 (EMB Global). Challenger banks and fintechs are not merely supplementing the market—they are capturing its core.

Shrinking Physical Presence

Major banks are closing branches at an unprecedented rate. NatWest closed 53 branches in 2025 alone, while HSBC’s digital transition has been described as “sluggish” (Finextra). The contraction of physical banking infrastructure is a global phenomenon.

Fintechs and Challenger Banks: Rapid Customer Acquisition

Trust Bank in Singapore became the country’s fourth-largest retail bank in just over a year, surpassing one million customers (The Straits Times). JPMorgan Chase’s UK digital platform reached 1.6 million customers in 2023 (Reuters). The revenue gap is widening: while legacy banks struggle for marginal deposit growth, digital-first banks are expanding at rates of 40% or more.

Micro-Transactions: The Global Bypass

The migration of micro-transactions away from traditional banks is most visible in rural and low-income markets, where agent networks and mobile money services—often run by telcos—have become the default. The Dominican Republic provides a concrete example of the scale of this bypass, even in the absence of a local fintech ecosystem.

• Population (mid-2024 est.): 11.5 million
• Rural share (official data): ≈ 85%
• Rural population using only feature phones (field surveys): ≈ 70%
• Active mobile phones (all types, global): 8.3 billion (BankMyCell, 2025)
• Feature (“button”) phones still in use (global): ≈ 1.1 billion (13%)
• Average fee for a sub-US $5 bank transfer in LICs: US $0.50–0.90 (10–18%) (MoneyTransfers)

Assuming each rural adult in the Dominican Republic makes one US $5 peer-to-peer transfer per day using a mobile wallet or agent network (with no bank involvement), and that a traditional domestic bank transfer of the same size would cost approximately US $0.60 in fees (mid-point of the 50 DOP Caribe-Express tariff):

• Rural users: 11.5 million × 85% ≈ 9.8 million
• Annual transfer flow: 9.8 million × $5 × 365 ≈ US $17.8 billion
• Forgone fee income: 9.8 million × $0.60 × 365 ≈ US $2.1 billion per year

Result: In this scenario, banks in the Dominican Republic forfeit approximately US $2 billion in fee revenue annually on micro-payments alone, while US $17–18 billion in transaction volume bypasses the formal banking rails. The same pattern is observable in many other countries with high rural populations and widespread mobile phone use.

Global Perspective: The Scale of Bypassed Banking

Extrapolating this model globally:

• Global rural population: ≈ 3.4 billion (World Bank)
• Assume 30% of those adults (≈ 1.0 billion) have access to a feature phone and mobile wallet/agent.
• Each makes one US $5 transfer per day.
• Average legacy-bank fee if using the formal system: 10% (US $0.50).
• Daily volume bypassing banks: 1.02 billion × $5 ≈ US $5.1 billion
• Annual volume bypassing banks: US $1.86 trillion
• Annual fee revenue forfeited: 10% × 1.86 trillion ≈ US $186 billion

Even if only one in four rural residents transacts daily and the average fee is halved, the annual loss remains between US $40–90 billion.

Structural Barriers to Reversal

• Physical reach: ATMs number just 2.95 million worldwide—a 2,800:1 ratio versus mobile phones. Branch networks are shrinking by 3–7% annually in most low-income countries.
• Cost structure: An on-us rural ATM transaction costs a bank approximately US $0.25–0.35 to service. Agent/mobile-money networks operate at less than US $0.08 per transaction, often subsidized by telcos.
• Device reality: Over a billion users have no smartphone; USSD/SMS wallets are their only digital finance channel. Banks rarely support USSD directly, while agent networks and telcos do.
• Behavioural lock-in: Once users trust a mobile wallet or agent for daily micro-transactions, larger remittances and savings migrate there as well, accelerating deposit leakage (already −4.8% in the U.S. legacy sector in 2024).

Lessons from M-Pesa: Missed Opportunities and Global Impact

A similar model has already demonstrated its transformative power elsewhere. In Kenya, M-Pesa—a mobile money service launched by Safaricom and Vodafone—now processes over $300 billion in transactions annually, with 51 million users and 59% of Kenya’s GDP flowing through its rails (Forbes, 2024). M-Pesa’s success was not the result of a bank-led initiative, but of a telco’s willingness to serve the unbanked and underbanked with simple, accessible technology.

Notably, Visa did not acquire M-Pesa. Instead, in 2022, M-Pesa and Visa partnered to launch the M-Pesa GlobalPay virtual Visa card, enabling M-Pesa users to make international online payments (Safaricom). The ownership of M-Pesa remains with Safaricom and Vodacom, who acquired the brand from Vodafone in 2020.

In 2013, a project with similar ambitions was proposed to banks in the Dominican Republic, aiming to bring agent-based, mobile-enabled financial services to rural populations. The proposal was dismissed as “excess service.” In contrast, M-Pesa’s model has since generated billions in revenue and transformed financial inclusion in multiple countries.

Conclusion: A Structural Shift, Not a Passing Trend

In countries with high rural populations and limited banking infrastructure, billions in annual fee income are bypassed on micro-transactions, with trillions in transaction volume moving outside the formal banking system. Globally, over $186 billion in micro-transaction revenue is now captured by telco-led or agent-based rails—roughly double the combined 2024 profits of the five largest U.S. banks.

The data is unambiguous: the migration of micro-transactions to mobile and agent networks is not a marginal trend, but a fundamental reordering of the financial landscape. The infrastructure and business models of traditional banks are being outpaced by the scale, efficiency, and reach of mobile-first and agent-based solutions—especially in rural and low-income markets.

Sources:

• BankMyCell: How Many Phones Are in the World? (2025)
• World Bank Rural Population Data (SP.RUR.TOTL.ZS)
• FinTech Weekly
• EMB Global: Digital Banking Market Size
• MoneyTransfers: Caribe Express Fees
• The Straits Times: Trust Bank Hits 1 Million Customers
• Reuters: JPMorgan Chase UK Digital Bank
• Finextra: HSBC Branch Closures
• Forbes: M-Pesa’s Global Impact
• Safaricom: M-Pesa GlobalPay Virtual Visa Card

For further context on the 2013 project offered to Dominican banks, see:

• Project details in Spanish
• Project details in Russian

The numbers, trends, and structural realities point to a new era in financial services—one where mobile and agent-based platforms are not just competing with, but overtaking, traditional banking infrastructure, especially in rural and low-income markets.

In November we wrote about how to “digitize” an apiary and turn it into “bee sharing”. In this article, its author, Artem Zharov, talks about the economics of the project and the change in business model.

https://pchelosharing.ru/
By Artem Zharov Founder and designer of BeeSharing and BeeSharing/Gift

Bees in the hive

Actually, the figure of 40 kopecks per year is not entirely accurate. To begin with, on average, a bee lives for 3-4 weeks, meaning each season sees 4-5 generations of bees working.

The 50,000 bees in a hive is also an average number during the peak cycle of the bee colony’s life. When summer begins, everything is in bloom, and the queen bee produces as many bees as possible.

…

“Only bees are better than honey”

Moreover, not all bees gather honey. Apart from the basic division of bees into the queen, drones (males, needed solely for fertilization), and worker bees, there is also a division based on roles, specifically by age. The bee’s body develops, and along with that development, its functions change. The list is ordered by increasing age:

• Cleaner bees
  .
• Builder bees
  .
• Nurse bees
  .
• Guard bees ‍
  .
• Ventilating bees
  .
• Forager bees (peak of bee development)
  .
• Scout bees
  .
• Water-carrying bees (old age of the bee)

But since all roles are necessary for production, we can calculate on average.

Bees can gather from 30 to 100 kg of honey per season depending on conditions. Obtaining 50+ kg is possible with an industrial approach, collecting nectar from agricultural fields.

This is 108% not our option because, at a minimum, we dislike monofloral honey (gathered from a single crop), and pesticides can easily find their way into honey.

…

Our bees gather around 40 kg, and half of this honey needs to be left for the bees for autumn, winter, and spring. So, we or the hive owners get a yield of 20-23 kg (last season resulted in 21 kg).

Industrial honey

The cost price of industrial honey is 20-40 ₽ per kilogram.

A kilogram of industrial honey in bulk costs 100-250 rubles.

In small wholesale, it goes up to 300 rubles, and at retail directly from the beekeeper, it can be up to 500 rubles.

At fairs, it can reach up to 700 rubles if purchased in 1-2 kg quantities.

In a store, after processing and packaging in a beautiful (or not so beautiful) jar, it can be around 1000-1200 rubles per kilogram.

…

Typical honey packaging is the complete opposite of our vibe.

The cost price of such honey is around 20-40 rubles per kilogram, depending on the arrangement of the apiary, its size, and most importantly, the yield.

All such apiaries use sugar, do not value the life of bee colonies, and do not hesitate to use industrially produced antibiotics, and so on.

…

COG of honey manufacturing.

If a beekeeper sells bee packages and other beekeeping products besides honey, having 500 bee colonies can generate a net income of up to 4 million ₽ per year. Blood money in the honey business.

…

Possible profit in an industrial apiary with 500 bee colonies: 4,000,000 per year.

Homegrown honey and honey from eco-friendly apiaries

Products like these are rarely sold in large wholesale quantities (several tons) because, with minimal sales skills, they can be sold at a good retail price. Also, there are rarely such large apiaries with such an approach.

Average cost price of homegrown honey: 200 ₽ per kilogram.

Summary of indicators and distinctive features:

• Yield: 30 to 50 kg depending on the region.
  .
• More work on the apiary.
  .
• Higher expenses for treatment and prevention.
  .
• No expenses for fuel, as the apiary is stationary.
  .
• Average cost price comes out to around 200 ₽ per kilogram.

…

COGS of eco-friendly apiary.

Apiaries that are now joining BeeSharing sell their honey in small wholesale for approximately 600-650 ₽ per kilogram. This often exceeds or equals their retail price. However, unlike retail, they don’t need to deal with:

• Packaging
  .
• Delivery
  .
• Marketing
  .
• Sales

Moreover, this is an expansion of the apiary without additional investments – hive acquisition is covered by us.

Thus, by selling through BeeSharing, adding a hundred hives results in an increase in the net profit of the existing apiary up to 1.6 million ₽.

…

1.6 million per year. This is an increase in the profit of an eco-apiary of 100 bee families working through bee sharing.

Working independently, without our platform, a beekeeper is forced to handle packaging, delivery, find clients, rent sales points at fairs or markets, advertise on platforms like Avito, maintain a blog…

Even if they sell at double the price, at 1200₽/kg (though discounts are common, especially for purchasing larger quantities like a 3L jar), the net profit comes out to roughly the same 650-700₽. Additionally, instead of staying at home with the kids and relaxing, they have to deal with sales.

…

Our additional values

We have our own apiary, and the figures above are partly based on it.

However, in addition to the apiary, we are developing a platform for beekeepers and a personal hive account for customers. So, in BeeSharing, apart from apiary costs, there are also the following expenses:

Content: It amounts to almost 350k per year, including:

1) A photographer takes photos and videos for photo reports published in the personal account. On average, 2-3 shoots per month.

…

2) A photographer takes photos and videos for photo reports published in the personal account. On average, 2-3 shoots per month.

…

Illustrations and design. It costs 250k per year. We create packaging for honey and related items. In our catalog, there are 50+ design templates, and each hive owner chooses how their yield will look as we approach autumn.

…

Including, we also create custom designs for clients. For example, here is a sketch of a label for the Cinemapicnic apiary of director Maxim Rusanov. The concept was approved on the first attempt:

…

Packaging and dispatch of orders + support — 600K per year. We package honey in St. Petersburg, not in the Krasnodar Krai where the apiary is located. From there, we dispatch parcels to customers using a courier service. Shipments usually reach Moscow within a day, and for St. Petersburg, we often provide same-day delivery.

…
New Year’s gifts for the employees of Pasha Molyanov’s “Let’s Do It” agency

Hosting, services, and such — 50K per year

Office rental — 480K per year. For packaging and for me. I have too many children to work from home

Communication at the apiary — 36K per year and a one-time (hopefully) 100K for equipment.

https://pchelosharing.ru/gift

And also: a discount of 9.99% with the promo code “СКОРО14И23”.

https://pchelosharing.ru/
By Artem Zharov Founder and designer of BeeSharing and BeeSharing/Gift

By Vladislav Vorkel
Product designer at Yandex.
Founder of the service for finding internships: https://t.me/Intern_Sheep.
Author’s blog: https://t.me/designvorkel

My internship search product currently has over 300 paying users, and it generates a monthly income equivalent to the salary of an average IT professional. I dreamed of creating a product that people would use, even if just a couple of them.

Hello, I’m Vlad Workel, a product designer at Yandex and the creator of InternSheep, a community with internships for designers. I’ll share how I really wanted my product and how I made it happen.

…

Roughly, this is how my dog helped in creating the product.

Internship search: where did the dream originated

I am a product designer and have worked in product roles, but I have always been interested in looking beyond the design field. I wanted to experience a product from all perspectives, from the idea, development, and demand to working with my team. Moreover, having my own product means it’s solely my responsibility, my decisions, and only my success or failure.

In simpler terms, it’s like crafting a wooden bow for a child that actually shoots, and then proudly running to show it to the adults.

The first attempt

About 1.5 years ago, I decided to embark on my first project – Gradeup. It’s a test for designers that, upon completion, allows you to assess your skill level and receive a competency map. It is still operational, and you can find it here: gradeup.vercel.app.

I found a developer acquaintance who was just learning front-end, designed the interface, and created a test consisting of competency cards from various companies. Initially, we tested the demand for the MVP test using Google Forms, where the result was a PDF that I personally drew and sent to each test-taker. I initially found the first users through my channel, which had about 500 people at that time.

There were many different hypotheses on how to monetize this test, but we never got around to it. Perhaps it was because there was a lack of understanding of how to do it, or maybe it was because I had already burned out and lost faith. That’s when the idea of gathering internships for designers came to mind, as it is something everyone needs!

The second attempt

Yes, that’s the one: InternSheep. So, the main story will be about testing the hypothesis, how we implemented the service in a couple of days with a developer in a hackathon format, how we attracted the first users, how the internship bot turned into a community, and how we created an educational program.

How not to start

The first week, I spent on absolutely unnecessary research, thinking I had to do it the “right” way. I scripted interviews, found respondents, and asked them about their job search experiences, problems faced, and so on.

The truth is, research is necessary when you don’t understand the target audience or when launching the product costs more than the research itself. In my case, I had a clear understanding of the target audience’s problems because I had personally gone through internships, and everyone around me was job hunting. It was literally part of my daily life. As a result, I abandoned the research and started building the product.

My opinion (it’s just mine, but I like it): in personal projects, it’s better to solve your own problem, then you won’t need extensive research. If your solution works for you, it will likely work for someone else – we are not unique.

…

The simplest hypothesis test:
There is a website where only internships are published. Internships of all cool companies immediately appear on the website after the company posts them on its official website. You will receive a notification. Are you ready to subscribe to the service for 200 rubles?

Yes: 52%
No, I don’t need internships: 18%
No, I’m sorry for the money: 17%
No, I’ll write in the comments why: 1%

From idea to the first revenue in just a few days

So let’s begin! The idea of implementation was to write parsers for career websites of top Russian product companies, we started with 15. Write a bot in which parsers will instantly send internships to users. They decided to sell the service by subscription – 200₽ per month

I assembled Landos in 1 day using nocode. I found several respondents for a decision interview and sold 3 subscriptions to a non-existent product. I just remember how I almost begged for real payment – only the money on the card could confirm the hypothesis in my case. It was a little awkward, but I overcame myself))

So, 600 rubles on the card, 0 developments, 3 people are waiting for the product within 2 weeks.

Let’s start development

I messaged Roman, a developer from Yandex, shared the idea with him, and we immediately started implementation. We needed parsers!

What parsers? Are you stupid or what?

Roman Dubinsky

Roman’s righteous laziness prevented us from making that mistake. Finding a student from Voronezh who browsed 15 websites several times a day turned out to be 100 times easier and cheaper. Later on, we did write parsers for some websites, but that’s not as important now.

…

Those very parsers…

In less than a week, we did the most essential thing: a bot and its admin panel where we uploaded internships.

The bot consisted of…

The bot consisted of:

1. Choice of Specialty: We had 5 options, but in practice, only designers subscribed.
   .
2. Subscription Payment: Done through the external Donate bot directly in Telegram.
   .
3. Each Internship had a link to the vacancy.

The entire flow seemed quite convoluted and makeshift initially, or at least, that’s what we thought at the beginning. In reality, it turned out that almost no one got confused, and people joined us smoothly.

So, the launch!

Actually, when I received the first prepayments, I started warming up the audience in my channel, talking about the service and collecting applications.

I asked all my acquaintances with channels to mention the service, and many responded absolutely for free. Thank you, if you’re reading this, it might not have worked out without you.

In the end, before the launch, we had over 300 applications.

As a result, we decided to sell it for 290 rubles to everyone who left a request, and then raise the price to 390 rubles.

We announced a broadcast where I presented the product and talked in more detail, but we started registering applications and opening access a few days before the launch.

…

The very broadcast where I tell you why you need to pay me 290₽.

At the time of the launch, we had sold about 200 subscriptions.

Couldn’t believe it when my card had not just 600₽ but 60,000₽. The amount may be modest, but it’s not about the money. I did something useful, and I’m getting paid for it! Wow!

Creating a community

Initially, we didn’t plan on doing anything like that. The idea was just to create a bot with internships – that’s it. But users suggested that it would be cool to have a chat where people in similar life situations could connect, receive support, and exchange experiences.

Both we and the participants saw significant value in this. Over time, it evolved into a community where we engage in various activities, communicate, and build friendships. People come to the product for internships but stay for the community. Many have found jobs or internships but continue to be a part of our community.

I was building a product and found friends.

What’s happening in the community now?

Continuous communication – and this is the most important aspect. People share test results, discuss who received invitations and who didn’t, seek advice, and find support when things get tough. Natasha, the community manager, and I have established a culture of zero toxicity and maximum friendliness. Throughout our time, we can count on one hand the instances of toxic conversations.

Lead designers conduct reviews several times a month where anyone can submit their portfolio or test work and receive a quality analysis.

We regularly host online meetups in our Safe Space, where you can come and simply talk when facing challenges at work or when despair sets in.

There’s also a book club and a poster art circle. Together, we read design-related books and discuss them, and in the art circle, we create posters with a designer from Yandex.Eats.

We are creating a small, warm, and friendly world where one feels inspired to grow and develop.

Periodically, members of our community would share internships that we didn’t initially have information about. We started keeping track of new companies, and as a result, the count increased from 16 to 26.

The product after 9 months

We haven’t closed down.
We are working!

Currently, the InternSheep channel has 2500 subscribers, with around 300 paying users in the product. Over 30 individuals have secured internships in top companies.

Throughout this period, I’ve hardly invested in advertising. There were trial sowings that yielded no results. Our growth is solely organic, which is great, but it’s time to learn how to attract more.

The team consists of three part-time individuals, including myself – a community manager/SMM, a designer, and me. Occasionally, we bring in others for one-time tasks or consultations.

What are our metrics?

As in any subscription-based service, the primary metric is retention.

We retrieve data from the Donate bot, which tracks all charges, subscriptions, and cancellations. We create charts that illustrate retention across monthly cohorts.

Currently, our average conversion to payment in the second month is 75%. I’m not sure how good of a metric that is, but it certainly seems quite favorable.

…

Retention from February to October.

I won’t talk about the number of paying users. It’s obviously important for us.

The second key metric is monthly revenue. From the start, we increased the cost twice – from 290 to 390, and then to 490. Subscription is not the only monetization model in the service – we sold partner products and channel advertising. Currently, we are developing our educational program and assisting companies in hiring talented interns, but I’ll write about that in the next article.

Let’s summarize

I dreamt of a product, and now I dream that it continues to live and grow. It’s a wonderful platform for experiments, for personal and professional growth. The most valuable thing I gained in the 11 months of working on the product is experience and passion.

Currently, the product brings in a monthly income comparable to that of an average IT specialist. A worthy reward.

Thank you for reading my story – perhaps it will inspire you to embark on your own pet project, or maybe it will just brighten your day a bit.

I would appreciate it if you subscribe to my Telegram channel (5000+) – I share stories about my projects, my work at Yandex, and the journey of self-discovery. I also throw in some humor occasionally.
The channel “InternSheep“: the product I talked about in this article.

By Vladislav Vorkel
Product designer at Yandex.
Founder of the service for finding internships: https://t.me/Intern_Sheep.
Author’s blog: https://t.me/designvorkel

If you, a respected Russian Apple user, do not want to switch to Android, carry cards or stick stickers on your favorite gadget, then there is only one way out – to pay via QR. In some countries, it is through “quaring” that most transactions take place. Today we will analyze how QR payments work and why their share is growing by leaps and bounds.

If you decide to pay for udon at a Shanghai diner with a card (even if that card is Union Pay), then it is very likely that Uncle Liao will make a surprised face and hand you a card with a QR code. Indeed, in China, almost all everyday payments are made by scanning black and white squares.

Instead of a price tag

Yes, China is the undisputed leader in QR payments. But many other countries are also actively getting used to them. Quaring is especially developed in Asia – for example, in Thailand, South Korea, India, Uzbekistan and other countries. But in other regions, it is far from dead and is actively growing. In Russia, QR-payments also feel good: the Russian Central Bank has developed them before as a feature of the Faster Payments System, and in the last year and a half, for obvious reasons, their share is completely flew up into the sky.

A variety of analytical offices (for example, this one) predict that in a couple of years more than 2 billion people around the world will use qar payments on a daily basis.

So, it is likely that payment via QR will become the dominant payment format in the near future, pushing cards into the background (I do not guarantee, but there is a big chance).

So, how does QR-ing work in general, what it is, and how it is better or worse than other methods. Let’s sort out.

1. Why does a QR code get along well with payments?

I would like to start with one simple thought:

A QR code, also known as a Quick-Response Code, is just some kind of information, but encoded as a set of black and white particles.

This information can be anything, for example:

• The history of some statue in the museum. Then we draw a kuar on the nameplate and glue it next to the pedestal.
  
• Information about your covid vaccination. In this case, the black and white ornament will appear in a special section on gosulugs.
  
• Or, for example, your bank account details.
  

In the latter case, we will talk about QR-payments (or, as it is sometimes called, “quaring”). But in absolutely all cases, the QR code will remain only a way to transfer information. In other words, QR payments are not some super-unique technology, but just the use of the most common QR codes in one particular area.

But it just so happened that the QR code has several valuable features at once, making it an ideal technological basis for a whole galaxy of new payment services. And here are the features:

• First, a QR code is an extremely undemanding thing. To display a QR code, you only need a screen, and to scan it, you only need a more or less tolerable camera. Simply put, a simple smartphone, which everyone has in the modern world, is quite enough. African tribes and Amish communes will be put out of brackets. However, no one forbids showing a QR code on a piece of paper. You can even attach a payment QR code to a cat – and then you get a payment QR cat (I hope you will forgive me this joke).

• Secondly, QR codes offer a huge number of combinations. The number of unique combinations of black and white squares is so titanic that I will not give it here. Just to mention that Chinese WeChat uses more than half a trillion (!) Kuars every year, and this is not even 0.001% of all possible combinations. In short, no matter how much the Chinese buy udon, we will have enough unique QR codes for a long time to come.
  
• Finally, over the past few years, everyone has learned to use QR codes. I think even grandmothers in the village have heard at least something about it (unlike these NFCs of yours).

Add to the above the fact that global payment systems (Visa, Mastercard and others) at one time forgot to cover the payment coir market, and you will get a suitable alternative tool for payments.

It is not surprising that in the last 5-15 years (depending on the country) it has been adopted by operators of national payment infrastructures and fintech services of digital ecosystems (more on this below).

2. What kind of payment QR codes are there?

So, although the QR code itself is an absolutely ingenious thing, it would not turn out to be called rocket science. Of course, there are also technological differences between the coirs – for example, the ways of encoding information (numeric, alphanumeric, byte, etc.).

But in this article, we will not dive deep into this – if you are interested, you can google 100500 technical articles about the technical nuances of these things.

Now we are talking about QR payments. And here, too, there are differences:

• The first is the source of QR. Who shows the QR code, and who scans – the seller or the buyer.
  
• The second is the saturation of the QR code with information. You can make a basic quarry containing bank account information. You can add some additional information to it – for example, the purpose of the payment. And also a specific amount, but here we are already moving on …
  
• … to point number three: the dynamism of the QR code. Perhaps this is the main technological difference. A QR code can be static or dynamic, and here we need to dwell in more detail.

A static QR simply contains the bank account details of the seller or buyer. In this case, after scanning, you will have to enter the amount yourself.

This is an unnecessary action for the client (i.e. complicating the client path, which is always not very good), but such a code can be pasted somewhere near the cash register and sending customers to pay with a slight movement of the hand.

Dynamic QR code is cool and advanced. But there is a nuance – if the seller wants to generate dynamic quarries, then you won’t get off with a cardboard box. You will have to buy additional equipment that can handle this very dynamic.

There are also several intermediate modifications. For example, when a static QR code leads to a dynamic web page. In this case, the QR code can be printed and not changed, but the payment chain will be extended by an extra load.

Or another option – a static QR contains a link to a dynamic page, from which, when scanning, the actual data is pulled into the payment application. As a result, a nipple system is obtained, when the QR can be left unchanged, but the buyer does not need to open an extra page either. True, the likelihood that something crookedly pulls up is growing.

A few words should be said about the source of quar. In the vast majority of cases, the QR is created by the seller (then it is Merchant presented QR), and the buyer reads it.

But there is also the reverse mechanics – the buyer opens the QR code of his bank account, and the seller scans it, thereby debiting the required amount. In my opinion, this is the most convenient way for the buyer (although not the safest – more on that below), but the seller will need a little more functional equipment.

3. Are QR Payments Perfect?

In fairness, it must be said about the shortcomings of quaring. They are like this:

• The first is “Internet addiction”. Card payments can exist without the Internet. If it does, then payment terminals can read data from a card or NFC, store it in their memory, and when they turn on the Internet, transfer it to the payment system. QR without the Internet turns into a pumpkin from the word “absolutely”. So, if your store is located in the basement with a poor network, then quartering may be contraindicated for you (the buyer will stupidly not be able to load the page after scanning).
  
• The second is: some security questions. The crooks are not asleep here either. Let’s say you show a QR with your account details sewn in at a small stall, the seller of which turns out to be a scoundrel. He quietly takes a picture of your code, and your money cried. Or another option – while the seller is gaping, the scammer will stick his scammer kuar over the real one. As a result, the money will not be spent on paying for your goods. However, scam is possible with bank cards. So, I will assume that quaring is not much inferior to him here (unless crooks use other methods).
  

• The third disadvantage is that QR payments require quite a lot of gestures. You need to take your phone out of your pocket or purse, then open the quaring application, then the camera will start to blunt during the scanning process… Then you will press the “translate” button, and the Internet will turn off, and the confirmation will hang. As a result, you stand in front of the cash register like a fool and think: “Have the money been written off, or do you need to do everything again?” It is because of such a hell of fine motor skills that many people initially spit on these coirs of yours and simply pay with a card.
  
• In addition, often there is no cashback for QR-ing, or it is significantly less than for cards. Such alignment, for example, in the Russian Federation. The reason is that the commission for QR is lower than for cards, which means that banks have nowhere to take budgets for cashback. However, this is not the case in all countries – in some, all fat cashbacks, on the contrary, are in QR payments (depending on how the size of commissions correlates in a particular country). However, this is not such a minus. Or rather, for whom it is like – a minus for the consumer, but a huge plus for business.
  

4. Why is QR-ing is especially preferred in Asia?

QR-payments are highly developed in many countries, but guys from Asia (including, of course, the main fans of quaring – the Chinese, as well as the founders of the technology: the Japanese) and residents of some other so-called. “developing” countries.

Why is that? I suppose that the explanation lies in the competitive situation in the markets during the years of the emergence and initial development of technology.

In the USA and Western Europe, by the beginning of the 21st century, card payments were already highly developed. Local banks actively issued Visa, Mastercard, AmEx and other cards, and the population was already accustomed to their daily use. Therefore, it was much more difficult for any exotic payment methods to squeeze out a share of the payment market.

But in all sorts of Thailand, Korea, Brazil and India, this was easier:

The penetration of cards is much less; for a significant proportion of the population, payment cards were still exotic at that time. Consequently, most merchants (perhaps, except for large chain stores) had no incentive to spend money on purchasing payment terminals. But everyone could put a sign with a QR code next to the cash register.

Then the classic “circles on the water” began. Business began to offer customers QR payment → The population began to taste it and slowly realize that the new payment method is generally not bad and works in many places → Demand for quaring from buyers has increased → Business began to introduce quaring even more actively. And so on in a circle.

It is worth noting that this spiral of growth would not have been possible without the active involvement of two important players:

• Firstly, national payment infrastructure operators (mainly based on local central banks). They found out that the method is quite working. So, you need to ride this wave. Well, you know, in order to reduce the share of cash (and, on the contrary, increase taxes), squeeze the market for all kinds of Visas with Mastercards …
  
  And in general, with the help of quaring, increase the share of domestic payments under your direct control. As a result, the states began to saw national payment QR services. Here is the Indian UPI, and the Brazilian PiX, and the Thai PromptPay, and, of course, our SBP. By the way, this logic may partially explain why quaring has become especially popular in countries with active state participation in the economy.
  
• Secondly, consumer ecosystems that have begun to actively adopt quaring. As you know, the core of any self-respecting ecosystem should be a payment service that enhances the flow of client activity between different products. You know, so that the user of an online cinema has a financial incentive to try food delivery or a taxi aggregator (or whatever else ecosystems usually have).
  
  As a result, in countries with strong national ecosystems, QR payments have become an important payment method. Here are the Korean KakaoPay, and the Kazakh Kaspi, and of course, the Chinese WeChat and Alipay.
  

Both “national” and “ecosystem” quarings had their own trump cards up their sleeves. State services set reduced acquiring commissions (much lower than card fees), funded from the budget.

Ecosystems, on the other hand, perfectly crossed quaring with their other advantages – for example, they accrued cashbacks that could be used in their other products. It was also easier to teach users of popular ecosystems to open the application for scanning – after all, the native interface is familiar and familiar.

As a result, QR further strengthened its position in the payment markets of these countries. Yes, if you do not take China into account, then kuaring is still difficult to call the main way. Rather the first among the alternatives. But its share is growing. And who knows what will happen next.

Friends, if the article came to you, then be sure to like it and send it to a friend who is fond of payments / fintech / banking.

Soon I will write the second part, where I will analyze in more detail the key QR services in different countries – SBP, Chinese WeChat and AliPay, Indian UPI, Thai PromptPay and many others.

Alexey Podkletnov

I analyze changes in business, industries and companies. More interesting things on my TG channel Disruptors.

Pavel Komarovsky
I write interesting things about finance at t.me/RationalAnswer

When Koreans made a sensational announcement about a new miracle material a few weeks ago, scientists around the world rushed to test it. And do you know who won this race? An anime girl from Twitter, who synthesized this wunderwaffe right in her kitchen in just a day, using the usual Soviet … (sorry, but that’s how it is!)

The heroine of the article smirks as if in response to all the haters from Twitter

If you missed all the hype at the beginning of August about the “new superconductor from Korean scientists”, then sit back, now we will explain everything to you. The intensity of the epic nature of this story in some places just rolls over, but first you have to dive a little into the background of the issue.

This article was co-authored with Mikhail Korobko, a quantum physicist at the University of Hamburg. As usual in such cases, Misha is responsible for the scientific outline of the story, and I am responsible for the creepy memes. Let’s Go!)

Superconductor: Why do you need this superconductivity at all?

Why is superconductivity at room temperature so interesting? And what is such a “super” in superconductors? Let’s sort it out:

When we run electricity through the ordinary wires, we always lose some energy. These losses are due to the presence of the “resistance” in the conductor: it literally resists to our attempts to push the electric current through it.

Amper here seems to want to say: “Help me, stepsister, I’m stuck!”. Oh, sorry, this is from another educational video…

The resistance comes from the physical structure of the conductor. Take, for example, a metal conductor: under the influence of a voltage applied to the metal, free electrons begin to move in it – this is what we call “electric current”. In the process of movement, electrons are “distracted” by the metal atoms themselves and interact with them, which, as it were, “slows down” the movement of electrons. As a result, the energy of the movement of electrons is converted into vibrations of the atoms of the metal conductor – in other words, the conductor heats up.

This process underlies, for example, incandescent light bulbs: the tungsten filament in them has high resistance and heats up very much when current passes through it, emitting bright light.

Strictly speaking, almost anything can become a light bulb if enough power is applied (see the fun raspberry nut)

But it would be nice not to lose extra energy on this resistance of yours where we don’t want it, right? Well, okay, in a light bulb, as we understand, the resistance is just in the subject. But otherwise … Here are just a few examples of scientific and technical goodies that could come into our lives if scientists came up with a new prodigy material with no resistance at all (that is, with superconductivity):

• Transmission of energy over long distances without loss. Built entire fields of solar panels in Africa – and provided electricity to all of Europe, easy-peasy!
• Efficient, cheap and powerful computer chips that use little power and don’t overheat – literally, a “supercomputer in your pocket”.
• And not only ordinary computers – with superconductivity it will probably be possible to finally cut down quantum computers (with the stability of which there are now big problems).
• In materials without resistance, interesting magnetic effects also arise, so you can throw all the levitating vehicles of the future and the ubiquitous super-high-speed magnetic trains here to the heap. Oh, and it will also be possible to make compact MRI machines (which now take up an entire room in hospitals), hang them in the subway, and read the thoughts of all passengers using neural networks!

Future with superconductors vs future without superconductors… Well, you understand: In a world of defeated electrical resistance, there will be happiness for everyone (and no one will leave offended!)

Brief history of practical superconductivity

At the beginning of the 20th century, scientists discovered that the resistance of some materials (such as lead and tin) drops to zero at very low temperatures (around 3 degrees above absolute zero: that is, at -270 degrees Celsius). These materials are called superconductors.

Over time, more and more different materials began to be discovered that showed the properties of superconductivity – but they all worked at hellish cryogenic temperatures (below the temperature of liquid nitrogen, about -196 degrees Celsius). Only at the end of the 80s did they find a class of materials that possess high-temperature superconductivity (well, “high” – a little higher than the temperature of this very liquid nitrogen).

– BatMobil works on superconductors?
– How did you guess?
– Uh.. Deduction and all that…

It would seem: what a problem, let’s just cool the wires with liquid nitrogen – it’s cheap and easy to use! But, unfortunately, almost all high-temperature superconductors are absolutely useless in terms of technology: they are brittle, difficult to process, and it is difficult to make any complex shapes out of them. Therefore, despite the discovery of superconductivity more than a hundred years ago, we practically do not find its application.

Well, okay, such “capricious” superconductors have found some very niche applications: they are used to make superconducting magnets in MRI machines and in all sorts of scientific installations such as the Large Hadron Collider, as well as in quantum computers (where superconducting elements are used as qubits). But in general, the revolution that superconductors promised us has not yet happened. In practice, for it, we need two conditions to be met: superconductivity at ordinary room temperature, and the practicality of creating and using such superconductors.

In recent years, several superconductors have just appeared at temperatures close to room temperature (in the region of -20 degrees Celsius) … But such materials become superconductors only under insanely huge pressure created between special diamond (!) Anvils. In general, again – far from practical application.

Typical news about the invention of the “room” superconductor looks something like this

So, we can say that superconductivity at room temperature and normal pressure is the real holy grail for the entire field of superconductivity – for 50 years since the discovery of high-temperature superconductivity, scientists have been struggling to create such a material.

It is curious that, in principle, there are no physical restrictions on the existence of such superconductors. But we do not yet have a complete theory of superconductivity so that we can “invent” them based on theoretical considerations.

The first theories of superconductivity arose only forty years after the discovery of the effect itself – in the early 50s. The most popular theory among scientists, which is still used today, is called the BCS (Bardeen-Cooper-Schrieffer theory) – it describes superconductivity at the quantum level.

Who do you think Sheldon Cooper from The Big Bang Theory was named after? You guessed it: in honor of this same Leon Cooper, one of the co-authors of the BCS theory!

This theory assumes that electrons in superconducting materials are combined into so-called Cooper pairs (we are talking about pairs here, not about vaping), which are all together in a single quantum state. In this state, they do not interact in any way with the atoms of the conductor, and therefore do not experience its resistance.

The BCS theory describes the main processes of superconductivity quite well, but still is not complete: after all, it cannot predict the specific properties of a superconductor and, moreover, does not allow inventing a material with the desired properties based on it. But the most important fail of this theory is that it is unable to predict high-temperature superconductivity – more precisely, it directly prohibits it! Currently, there are several theories that are trying to replace the BCS – but so far none of them can fully explain superconductivity for an arbitrary material: different theories work better for some materials, but worse for others.

In such conditions, scientists have to look for new materials almost blindly, using the so-called “scientific poke”. From time to time, someone throws out a new loud announcement on the Internet about the long-awaited discovery of a room temperature superconductor (here are just a few examples: one, two, three, four, five), but usually no one even pays much attention to them: after all, they almost always turn out to be either are not reproducible by independent scientists, or do not even receive publication in scientific journals at all.

Not so long ago, a big drama broke out with the withdrawal of an article from the prestigious journal Nature about another superconductor at room temperature – it turned out that the data there was simply stupidly fabricated. Well, there is a scientist himself named Ranga Diaz, it must be admitted – a very specific character, and who has already become famous thanks to his “skillful” fakes…

The same Ranga Diaz says in an interview: “I’m sitting on a fishing trip – I’m biting … I’m hooking, and there’s a superconductor, diamagnetism in the entire field, electrons in it just rustle – you can see it with the naked eye !!”

Korean sensation and flying stone

In short, we are already coming close to, in fact, the history of the superconducting superhype of recent weeks: at the end of July, two preprints from Korean scientists appeared on the arXiv website, one of which bore the “modest” name The First Room-Temperature Ambient-Pressure Superconductor. We must immediately make a reservation that arXiv is not a peer-reviewed scientific journal – there is a certain moderation there, but it works rather “by external signs” and cuts off only very obvious pseudoscience and bullshit.

Even when you read the abstract of the posted preprint, it becomes obvious: with one hand the author was typing letters on the keyboard, and with the other he was frantically imagining how he was already receiving the Nobel Prize for the amazing foundations of being discovery

To say that the whole Twitter bombed from this news is to say nothing. As we wrote above, usually superconductivity does not arouse any special public interest; what exactly attracted tens of thousands of people in social networks this time is hard to say. Perhaps it’s the video demonstration with proof of levitation in the received LK-99 sample (that’s how the Koreans called their discovery) that accompanied these two preprints; or maybe that at first glance there was no obvious crap in these two papers, and serious scientists reacted to them in the style of “hmmm, curious, well, let’s figure it out!”.

A copy of the very original vidos from the Koreans. A little later, they handed over another version of the video to The New York Times – there was already a thermometer in the frame to prove the “room” temperature.

What is interesting about this video, and what does levitation have to do with it? The fact is that in addition to zero resistance, superconductors also have an interesting property: they levitate in a magnetic field – the so-called Meissner effect. (We will not go deep into the details of the physical process here, if you are interested – you can look at this recent article on Habr.)

It was this levitation that was demonstrated in the original video, and it was also posted by various teams of scientists and amateurs in the following days. The message was often this (especially in the interpretation of non-specialists): if it levitates, it means that it is a superconductor! But, strictly speaking, this is not necessarily the case. Magnetic levitation is possible for other materials, and even for living beings: the respected scientist Andre Geim once received not only the Nobel Prize for the discovery of graphene, but also the slightly less prestigious Shnobel Prize for experiments on the levitation of a living frog.

The frog of Andre Game starts an experiment on magnetic levitation, I see this (according to the rumours, not a single amphibian was harmed in the process)

This effect of diamagnetism is very similar to the Meissner effect, although it requires much stronger magnetic fields. However, levitation is the simplest first sign of possible superconductivity; and, in fact, the only one available for testing without specialized equipment. Other confirmatory measurements (say, zero resistance of direct interest to us) require great accuracy and special methodology. And levitation, besides, also looks cool!

In general, the whole world held its breath somewhere at this moment: will there be independent confirmation of the Meissner effect – or will everything, like in previous times, stop at the stage of too loud and too hasty statements from the Korean team of scientists? It is here that the heroine of our story enters the stage …

“That Russian Anime Profile Pic Girl” comes into play

While ordinary people on Twitter were talking about how many weeks (or months) it would take for the world’s leading laboratories to replicate LK-99 and confirm its superconductivity, a little-known @iris_IGB account with an anime girl in the avatar decided to take the initiative into their own hands.

To summarize the essence of shieldposting from Iris, it boiled down to something like this: “You are all stupid and not being treated; and if you had at least a little understanding of physics and chemistry at the level of an advanced Soviet schoolchild, you could easily synthesize the right substance yourself!” (I’m exaggerating a little – but, to be honest, not too much).

Iris’ most epic viral thread literally starts with “my girlfriend has a carpet cleaning today and we can’t watch Kin-dza-dza together… so it’s time for room temperature superconductors!” This is followed by an extremely informal description of the theoretical calculations on the theory of superconductivity, and after them – a step-by-step process for obtaining an analogue of LK-99 with pictures in less than a day.

Scraps of Jacobs coffee and creative shit in the kitchen – as some tweeters comment, “this is the atmosphere in which real science from catgirls is going on!” (important disclaimer: the catgirl is Iris’ friend, not herself)

Several things added to the epicness of everything that happened at once:

• Firstly, Alexandra Iris does not specialize in superconductivity physics at all, her area of expertise is soil molecular biology (!).
• Secondly, as she writes herself, Iris did not aim to replicate the entire process from the original paper of the Koreans – she immediately tried to improve their approach, based on her understanding of the likely nature of the observed effects.
• And thirdly: it seems as if she succeeded! In her posted slightly jackal photos, a small grain of some material (“speckle of shit,” as the author writes) is visible, which “levitates” above a magnet inside a hollow tube.

July 30th tweet, 3.7M views: First independent photographic evidence of LK-99-like properties on the entire web! (Unless you’re into anime, “fanservice” is usually used to refer to erotic content for die-hard fans of a series.)

Then everything developed exactly as is customary on the Internet: they began to show Alexandra for the dumb quality of the photo and for not wanting to record a video – she snapped in response in the style of “and you open any Soviet textbook, synthesize the same thing using it – and shoot your own fanservice in whatever format you want!!”

Regular references to the USSR in Iris’ tweets are not accidental here: she herself firmly stands on the position that the Soviet scientific school simply hides, without getting up from the couch, everything else that is happening in scientific laboratories around the world. And all this mess with the LK-99, they say, is an ideal illustration of this!

After all, Lee and Kim (the authors of LK-99 – in fact, it was their names that formed the title of the material) in the 1990s worked with the Korean professor Tong-Seek Chair (Choi?), who had previously worked together with Professor Galashevich in Poland – and he, in turn, was a student of the Soviet scientist Nikolai Bogolyubov, who in the 1950s created his own theory of superconductivity. In the scientific mainstream, this theory did not take root very much, losing the battle for the minds of this most popular Bardeen-Cooper-Schrieffer theory, but … Perhaps there was something in it after all? At least, it seems that this is what Alexandra Iris herself thinks.

Prof. Bogolyubov, judging by the archival photos, was a rather stylish guy. I would venture to suggest that if he had been born in the early 2000s, he would now be a regular at anime festivals …

Epilogue: it seems that the miracle did not happen

Immediately after @iris_IGB, other independent checks began to slowly come in. Somewhere the samples levitated, somewhere they did not; but each new video and photo was met with another wave of delight. Serious scientists reacted to them, trying to explain that these observations do not yet prove anything – such reactions also went viral, and the public was again disappointed.

The swing between delight and disappointment has led to the fact that the story began to be followed far from the scientific community – like a cool series, stocking up on popcorn. Adding fuel to the fire was the fact that the original preprints themselves came out with drama: the co-authors accused each other of violating scientific ethics – they said that the articles were posted without the permission of the co-authors, quietly. At the same time, the preprints themselves contained visible errors in the graphs and a lot of inaccuracies, which somewhat complicated the replication process (go figure it out – did you get exactly what you need, or some other material?).

Soon, the pebbles were already “flying” in several laboratories, and everyone completely did not care about all the problems of preprints. A week later, theoretical articles were published with numerical calculations of the structure of the material, which turned out to be close to that expected from a superconductor. Although many other materials could have such a structure, the new burst of hype was unstoppable: not only did the stones fly, but the “theory proved everything” already!!

But then everything went a little downhill: last week more serious studies of a new material began to appear, and so far none of them shows the desired properties of a superconductor.

Yes, it flies in a magnetic field – but just like an ordinary ferromagnet. If in earlier articles it was possible to say that the material was “wrong”, “it was not synthesized in the right way”, then here the structure has already been confirmed with accuracy.

In one very recent study, scientists made pure LK-99 without any impurities – and it turned out to be not only not a superconductor, but also an excellent insulator! And all the properties that made it look like a superconductor were actually due to impurities of copper sulfide that got into the manufacturing process (why these impurities have such an effect is explained in a separate article).

In general, so far, of the many attempts to reproduce observations, not one has shown superconductivity. There is a feeling that this opening can be closed – at the moment there is a consensus that there is no smell of superconductivity in LK-99. Of course, we can wait for more independent reviews, publication of results in peer-reviewed articles and other things – but there are still not so many reasons for optimism…

On Twitter, LK-99 has been buried with honors for a whole week.

However, Alexandra Iris herself does not seem to lose her optimism: she continues to tweet reasoning about her thoughts about LK-99, and also trolls her critics with the help of rebukes stylized as visual anime novels. If I understand correctly, she believes that LK-99 itself is not a superconductor; however, she says that she observed the Meissner effect in the grain synthesized by her with her own eyes and … I don’t know, here I’m already confused what specific conclusions should be drawn from this – but personally I don’t exclude that in six months we will see a new interesting a scientific article from some Moscow research institute.

In the end, on the selfie published by Alexandra, something like this happens – enthusiastic people are sawing some kind of prodigy in the laboratory …

Sometimes you still really want to believe that people with anime avatars, who have obscura knowledge of the ancient Soviet sages, are able to save the world – right?

PS: If this article gets a lot of likes, Mikhail Korobko and I will try to make a follow-up to it with comments from Alexandra Iris herself, as well as with the opinion of some specialist physicist directly in the field of superconductivity. If you don’t want to miss the following materials on the topic, then we invite you to subscribe to the TG channels of the authors: Mikhail Korobko’s Homeostatic Universe (for those who want to fumble for physics and science) and Pavel Komarovsky’s RationalAnswer (for those who are for a rational approach to life, but prefers a little easier).

Hello! My name is Vlad Proshinsky, I have been launching and developing IT products for 8 years in the role from Product Manager to CPO. For the last 8 months I have been studying how AI and LLM can be applied in products and business processes. I am convinced that knowledge in the field of neural networks will give managers and founders a great advantage

In this article, I want to share my strategic vision, cases of using generative AI in products and in automating business processes of world-class companies.

Generative AI is a type of AI that creates text, photo, video, audio and other content at scale using data arrays (LLM) and machine learning (ML)

Generative AI has had an impact like no other technology in 10 years. For six months, ChatGPT managed to prove that neither the cost of computing, nor the complexity of the infrastructure, nor human resources are no longer an obstacle to mass adoption, the AI transformation has begun.

According to a McKinsey report, AI could generate an additional $13 trillion in growth. by 2030, increasing global GDP by 1.2% per year.

Domination and strategic context

Mass adoption of technology in the active phase, not only in personal use, but also in business processes. The productivity gains of developers and knowledge workers using GPT-based products are massively proven, and have forced organizations to rethink their business processes and the value of human resources. The way product solutions based on GPT and other neural networks develop, how AI is integrated into everyday life of a person and business is another proof that AI is a strategic necessity.

Source: Bloomberg News, income statement analysis. Note: analysis in the context of the terms “AI” and “Artificial Intelligence”. Data for July 27, the share of Nasdaq100 informing companies was 40%

In this quarter’s earnings report, US tech CEOs are talking less about the problems and more about how they plan to use AI to boost revenue and lower costs. Mentions of “AI” have increased more than 4 (!) times compared to 2022, indicating that AI has become a key driver of competitiveness and innovation in the technology industry.

The AI winners take it all

“Winner takes all”, and in the context of AI, this expression has a direct connotation. In the technology race, leaders are afraid of being left behind if they slow down and others don’t. This “prisoner’s dilemma” from game theory creates risks for responsible AI practices. Speed-to-market leaders are driving the current “arms race” in which big corporate players are rushing to release neural network-based products.

CEOs and founders are actively pushing teams to find ways to incorporate generative AI into their products and processes, but to create value with these technologies, you need to understand how they work, where they will bring benefits, what their limitations and risks are. Such competencies are usually rare among top management. Therefore, they need to be looked for on the side, on platforms like CareerPaths or freelance platfroms like Upwork or Fiverr.

In the article, I will briefly touch on the development of technology, as a product specialist, I am interested in how this technology will change the needs of users, ways of solving problems, and most importantly: “how to make money on it?”, After all, the task of a business is to make a profit. But according to statistics, 87% of ML projects do not even reach implementation, and the main problem is that few people now have comprehensive knowledge about what kind of “cube” in the form of an AI solution to build into the system, how to do it, how everything is not “ break”.

Common problems teams face:

• Wrong processes of workflow on an AI project
• Misidentifying Opportunities for AI Applications
• No vision (experience) → a lot of resources are wasted
• Business and products still do not speak the language of DS and ML engineers
• Do not identify possible risks at the start
• Passion for experiments, ignoring business goals

In today’s realities, businesses need managers who will see strategic opportunities to implement AI where possible, as well as build operational end-to-end processes.

Strategic vector

Strategic innovations that require special attention (2 to 5 years to mainstream adoption) include generative AI and decision intelligence. Early adoption of these innovations will lead to a significant competitive advantage and will simplify the challenges associated with using AI models in business processes.

Gartner

According to the graph provided by Gartner, general AI (AGI, artificial general intelligence), also known as “artificial intelligence”, capable of finding a solution for arbitrary tasks and learning to infinity, will appear in about 10 years. The forecast of the Metaculus community is slightly more optimistic – January 2027.

Obviously, development progress towards AGI is artificially hindered: the first letter demanding to immediately suspend training of more powerful models than GPT-4 was published in March, the second – in May. As a result, in mid-April, OpenAI announced that they had suspended the training of the GPT-5 model and were engaged in “other things” (fast API and reducing the “AI cost”), and on July 30, the same OpenAI applies for the GPT-5 patent… Process irreversible.

An intermediate step is “causal / casual AI” (Casual AI) is a new type of AI that can detect and reason about the relationship of causes and effects, as well as determine the strength of each connection. Causality is very important for subsequent progress. It is characterized by: high autonomy, adaptability, making informed and logical decisions.

Causal AI is a key stage in the development of AI towards greater decision-making automation, autonomy, reliability and common sense

Gartner

On July 14, 2023, Elon Musk announced the launch of xAI, the goal of which is to “understand the true nature of the Universe,” and Casual AI is exactly what is needed to solve such problems. Among the early tools: Auto-GPT, BabyAGI, SuperAGI, ToddlerAGI.

Current (operational) context

Now in most AI products under the hood is GPT-3.5 or GPT-4, which is further trained to solve specific problems within the product, and the most popular product is ChatGPT, which gained 1 million users in a record 5 days in 2022.

The appearance of GPT-5 will undoubtedly improve the quality of products and open up new opportunities for solving problems and ChatGPT itself. Thus, your GPT assistant, which can help you do your work more efficiently, will become smarter over time, and the quality and speed of tasks will grow exponentially.

Companies and professionals who have implemented AI tools in their work win in the moment and over time, the gap from those who do not use them will rapidly lose to more advanced ones.

Why is this so important right now?

A report by Microsoft researchers explains how the GPT-4 language model learns human thinking skills. Can this be considered a milestone in the emergence of AGI (Artificial General Intelligence)?

Given the breadth and depth of GPT-4’s capabilities, we think it would be reasonable to view it as an early but still incomplete version of General AI (AGI).

GPT-4 has signs of human thinking, the argument is that GPT-4 achieves human-level performance in many tasks.

S´ebastien Bubeck (Microsoft Research)

ChatGPT can be considered as a mediocre assistant, the level of which depends on the skills of prompting (the ability to communicate with neural networks in the form of prompts), but now it is the starting point for a completely different future, which will come in 3-5 years.

Product changes

Global product companies are actively implementing AI into their products, adding value by increasing the speed of work and solving problems from scratch, this allows you to better activate users in key JTBD scenarios. Examples:

• Zoho added ChatGPT in Notebook
• Taboola planned to add ChatGPT to your advertising platform
• Microsoft added ChatGPT in Bing and Edge
• Google implemented AI in Docs, Sheets
• Miro released an AI launchpad (as they did in Notion earlier)
• UIzard added ChatGPT to generate landing pages and UI applications
• BuzzFeed planned use ChatGPT to create quizzes

AI Tools

A small list of tools that you can start using in your work and in the company:

Marketing

• search for hype, trends, insights: AnswerThePublic – listens to autocomplete data from Google queries. Neil Patel says: Users share their deepest secrets with the search engine, which creates value for companies that aim to get to know their customers better.
• create articles and posts up to 10 times faster: Copy.ai, CoWriter, Jasper, Phrasee, AI Writer, WriteSonic, HyperWrite, GetGenie Ai)
• hyper-personalization: Persado has developed “motivational AI,” as the service claims: «Brands using Persado AI are seeing a 41% increase in conversions across digital channels». Their algorithm is focused on achieving a positive ROI
• creating banners and creatives: AdCreative.AI, Pebblely, Fotor, Designer
• creating sales videos: Designs.AI, Kaiber, Elai.io
• competitor website monitoring: Browse AI 
• LP and sites: Landingsite.ai, Framer, Uizard , Durable, SiteGPT, 60secSite
• creating realistic people with your product: Caspa
• SEO: Craftly.AI, SEO.app, Charm, SEMRush, Moz, Market Brew
• AI for email: Animaker, Postcards 
• presentations: Pitchbob, Tome, Designs, AI Colors, Gamma, Pitchyouridea

Sales

• sales forecasting: RCMX
• train salespeople on effective models: Fiber AI
• digital sellers: Rask.AI, Immersive Fox, Veed AI, Elai.io
• call analysis, translation from audio to text: Fabius, SalesAI
• autocomplete customer cards: SalesAI

Product

• look for growth points in a product for products: Kraftful
• user retention: Subsets
• sort user issues: AppFollow, Charm
• segment users: Obviously AI
• generate growth hypotheses: Goal AI
• analyze customer feedback: Rargus
• conduct user research: Outset.ai, Iris.ai
• conduct market analysis: Silatus, Pitchbob.io
• Prompt engineering for products: AI Product

Support

Forbes writes that Meta , Canva , and Shopify use ChatGPT-based solutions in their customer support chatbots. Support Solutions: Kapa , Clueso , Ada , MagicReply

Development

According to GitHub, Co-pilot is involved in 46% of the code (among those who use it) and helps to write code 55% faster

• business process automation: MagicFlow , Bardeen , Levity.ai
• assistants: Co-pilot , Sweep , CodeStory , Autocode , GPT Engineer , Tabnine
• code review: AI Code Reviewer , Backend-GPT , Planar
• creating LLM language models: Flowise , GradientJ
• code auto-documentation: Stenography , Mintlify
• Cody – helps you read, write and understand code
• Giskard – helps reduce performance errors in AI models
• Lightning – helps train, deploy and build AI
• Tempo – writes the front-end
• AutoInfra – DevOps and Infra automation
• Structured – parsing logs

Finance

• budget optimization and forecasting: FinanceGPT , OpenOs , ObviousAI
• work with data in the “request-response” format: Code Interpreter , Seek AI
• chat with PDF: AskYourPDF , ChatDOC , ChatPDF , Fileread AI
• database analysis: Loofi , Baselit, Noteable (SQL, Python, & R)
• AI training on company data: Fine-tuner , ChatNode , Fini
• financial assistant: Quill AI
• workflow: Deep Cognition , Acodis , Mindee , Smart Touch

HR / recruitment

• automation of the recruitment process (job sites → resume analysis with ChatGPT)
• layoff forecasting ( Xsolla case )
• resume preparation: HyreSnap , Resume.io
• Interview Coach: Yoodli

Management / Management

• Kona – Executive coaching
• Stellar – sets goals / KPIs in the company and monitors their implementation

It is important to note the nuances of working with GPT:

• “Garbage in → garbage out” is a clear description of ChatGPT. First, master the skills of writing prompts ( cheat sheet ).
• Hints and questions lead to better results. If ChatGPT does not know the answer, it will invent one (needs to be double-checked) and can forge sources and give plausible but incorrect or nonsensical answers
• the accuracy (accuracy) of the answers may be reduced. Study of accuracy reduction from 52% to 10%

In summary, I want to note that at the stage of business growth, experiments play a special role. Testing hypotheses is one of the main tasks of marketers and product developers. By testing the listed tools, there is a high probability that you will be able to find points of increase in efficiency in your work.

***

The effect of the introduction of Synthesia AI avatars , as examples of growth that we plan to implement in one of the client’s projects:

“Thanks to AI avatars, we had 35% more meetings than at the last exhibition

Infinite Peripherals / VC of Marketing

The training team cut 50% of the cost of voice over video compared to hiring multilingual professionals

Xerox / Global L&D Sales Learning Strategy Manager

We reduced video production time by 95% from weeks to hours.

Network Rail L&D / Multimedia Development Manager

We increased engagement in online learning by 30% compared to text modules

BSH / Global Supply Chain Training

Open source AI

Separately, I will highlight open source solutions. They are more suitable for medium-sized businesses, for projects that require more flexibility, security, and budget savings.

According to Trends.vc analytics , open source AI is helping companies build infrastructure faster by learning from each other. Companies that offer closed source AI decide when and what you can use it for. Open source AI helps us learn from and build on each other’s work, it turns the AI ​​arms race into collaboration.

Open source platforms

• Huggingface – creating and deploying AI Open-source models
• Replicate – create and run AI models in the cloud
• Google Colab – for machine learning research
• Difty – open-source for LLMOps, AI-native applications

In the future, there will be more platforms built to host open source models. They will simplify the creation and deployment of models.

Open source models

• Panda GPT – will create a description from the picture, write stories, videos and answer questions from the audio. Can simultaneously accept multimodal inputs and compose their semantics
• Stable Diffusion – transforms text into an image (text-to-image)
• Riffusion – converts text to audio (text-to-audio)
• BLOOM – open multilingual model
• DragGAN – changes the poses, shape, expression and location of objects
• Falcon – LLM for Research and Commercial Use

Open-source datasets

• The Pile is a dataset of books, web pages, chat logs, and more.
• ImageNet – 14,000,000+ images with comments
• OIG – Conversational Data for AI Chatbots

open source tools

• PyTorch – a framework for creating deep learning models
• TensorFlow – open source ML platform
• Keras – API for deep learning models
• Prem AI – LLM Deployment Tool on Servers

Open Source Alternatives to ChatGPT

• GPT4All – free local chatbot
• HuggingChat – ChatGPT alternative based on Open Assistant
• StableVicuna is a chatbot that learns from people’s feedback

Advantages of open-source AI

• Open source solves vendor lock-in and high shift costs. Platforms like Hugging Face make it easy to find the right AI models for your options.
• Open source AI allows you to create niche applications that large companies (like OpenAI) with closed source do not have the time, knowledge or interest to create.
• Clear documentation improves the quality and adoption of your open source AI tool.
• Open source AI models may be slightly worse, but much cheaper.

Opportunities for Entrepreneurs

The emergence of a new “job” (in terms of JTBD ) in consumer segments is already actively taking shape. This generates demand for expertise in implementing AI into business processes to gain competitive advantage. In this regard, AI consulting will be actively developed , as well as courses and master classes on implementing solutions into business processes.

Companies will increasingly outsource such tasks as there is a high risk of internal sabotage among employees due to fears of being out of work (a 2023 WEF report states that by 2025, 50% of workplace tasks will be automated).

With the formation of a market for AI business consultants, launches of info products will gain momentum. You can create courses and master classes to implement AI in business processes, as well as train freelancers to master the tools. For example, English-language master classes:

• ChatGPT for Salesforce teaches you how to build ChatGPT for Salesforce
• ChatGPT Master Class on how to start and scale a business
• ChatGPT course for youtubers how to find ideas, write scripts
• ChatGPT: The Ultimate ChatGPT Copywriting Class teaches copywriting
• AI Assistants Teaches ChatGPT Marketing, Business, Finance
• ChatGPT Zero to Hero teaches you how to become a ChatGPT expert
• AI language learning revolution how to create classes with ChatGPT
• ChatGPT for Work teaches ChatGPT writing, development, translation
• Take your business to the next level of content marketing with ChatGPT
• ChatGPT Secrets To Master Data Science teaches data science in Python with ChatGPT

Summary

Generative AI is at the peak of the hype, according to a recent Gatner report, expectations from the technology are overstated and, apparently, we are moving into a phase of “disillusionment”, as evidenced by the decline in the use of ChatGPT among enthusiasts (in the Russian Federation, the ChatGPT audience decreased by 30%, and Midjourney – by 50 % in June) , however, according to the cycle and forecasts of strategists, the technology will be widely used in early 2024.

Recently, Gartner conducted a survey at The Future of Data-Centric AI conference , which showed that the majority of participants planned to enter the industrial use of large language models (LLM) by the end of 2023. Companies are forced to add AI to their decisions and processes in order to remain competitive.

In this article, we will tell you about the most important features implemented in ChatGPT over the last six months (the most powerful neural network in the world). Additionally, we will discuss the vision of the future shared by Sam Altman at the OpenAI conference held on November 7th. Spoiler alert: they want to create "Smith agents" that can independently interact with the world!

…

Sam Altman here be like, “Well, we’re basically going to train these agents and release them into the network – as for what happens next, just watch the Wachowskis’ movie, I won’t spoil it for you…”

This article appears to have two authors, but in reality, almost the entire text was written by Igor Kotenkov (the author of the Sioloshnaya channel on machine learning, space, and technology). One could say that Igor was responsible for technical accuracy and expertise in artificial intelligence. After that, Pavel Komarovskiy (the author of the RationalAnswer channel on a rational approach to life and finances) piled on top with some quirky memes. In short, no time to explain, let's go!

Since the release of our previous article, “GPT-4: What the New Neural Network Learned and Why It’s a Bit Creepy,” a lot of interesting things have happened. There have been updates to existing products as well as the release of entirely new ones.

Developers are racing to create new AI startups, companies are attracting billions of dollars in investments, and people are getting lost in the news, struggling to understand what’s happening in the world of artificial intelligence. In short, we decided it’s time to provide an overview of the key changes that have occurred over the past six months and share the latest announcements from the just-concluded OpenAI DevDay 2023 conference. Even if you’ve been closely following the development of ChatGPT, we’re confident that you’ll find it informative and interesting!

…

Note on ChatGPT/ChatGPT-3.5/GPT-4 to avoid confusion (read only for sticklers and pedants):

In general, all these terms roughly mean the same thing. But let’s clarify the terminology we use:

• LLM, Large Language Model — a large language model. Basically, any text neural network, with ChatGPT being a prominent representative.
  GPT-3.5 — the basic text model (LLM) from OpenAI, which existed for a long time as a service for developers. In terms of capabilities, it’s similar to the version that went viral in December 2022, known as ChatGPT.
  .
• ChatGPT, also known as ChatGPT-3.5 — the first version of a conversational AI assistant based on GPT-3.5. Dialogue format was added, and specific training was conducted for this format.
  .
• GPT-4 or ChatGPT-4 — an advanced version of the model from OpenAI. It’s larger, trained for a longer period, making it smarter and capable of understanding more languages. It was added to the ChatGPT website immediately, so effectively, since March 2023, ChatGPT can denote GPT-4: the terms are used interchangeably. A separate version of GPT-4 without the chat format has never been shown to the public.
  .
• In essence, ChatGPT refers to the conversational LLM in general. In almost all contexts, it can be perceived as GPT-4, as there is no point in discussing older and less capable models. So yes, ChatGPT = GPT-4.

…

If you haven’t read our two previous longreads explaining in simple language the principles of how text neural network technology works, now is the time to catch up (it will help you understand the current article):

• How ChatGPT Works: Explaining the Evolution of Language Models from T9 to Marvel
  .
• GPT-4: What the New Neural Network Learned and Why It’s a Bit Creepy
  ChatGPT is Making Waves Across the Planet

ChatGPT is sweeping the planet

First, let’s say a few words about how ChatGPT has evolved in terms of popularity and penetration into the masses. (By the way, a survey among the authors of this article showed that 50% of them regularly use this neural network!)

Sam Altman (CEO of OpenAI) at the OpenAI DevDay 2023 conference revealed the following statistics: the Weekly Active Users (WAU) of ChatGPT exceed one hundred million people. Interestingly, the weekly metric is not the most commonly used; usually, people talk about Daily Active Users (DAU) or Monthly Active Users (MAU).

We remember that at the beginning of 2023, more than 100 million people were already using the product monthly. Let’s cautiously assume that this figure hasn’t dramatically increased, so it was decided to slightly change the presentation. According to internet traffic calculations, MAU is approximately 180 million people, which is still very impressive for a year-old product!

If you’re a finance enthusiast, the following should catch your interest: 92% of companies on the Fortune 500 list (the largest U.S. companies by revenue) are already using OpenAI products. In short, businesses are actively trying to figure out how to make the most of this technological singularity to earn more profits!

…

However, as they say, there’s a catch: if you train a neural network on a database of chats where programmers communicate about work, it quickly starts responding to any queries in a style like, “Oh, I’m feeling burned out, I could really go for a fresh smoothie right now…”

And most importantly, all of this has been achieved entirely without any paid advertising—just the product, with word of mouth spreading the news! (Disclaimer: this material has not been paid for by OpenAI).

Okay, now let’s briefly go through the key milestones in the development of OpenAI’s brainchild that we’ve observed since the release of the flagship GPT-4 model in March 2023.

Spring 2023: Tools and plugins for ChatGPT, or how to add “handles” to a neural network

Many users have long and rightfully criticized the “limited” capabilities of language models since they don’t have access to the internet—meaning they cannot find and use fresh information to form responses to queries.

All the knowledge they possess is dictated by the training data the model has seen. Moreover, in their original form, Large Language Models (LLMs) are not particularly strong in mathematics, performing only approximate calculations (though sometimes they may be accurate).

Recognizing this limitation, OpenAI adapted the concept of “tools.” Just as a person uses a calculator for complex calculations instead of mental estimation, ChatGPT can turn to an external service to perform a specific action—even if it’s much more complicated than adding two and two. Shortly after the release of the GPT-4 model, “plugins” emerged, with the main ones being access to the Bing search engine (oh, no jokes about the model “googling”!) and a code interpreter.

The first helps update knowledge on various topics by passing the results of the search engine’s work on a specific text query (which the model itself formulates) to GPT. The second determines when the model wants to run a Python program, performs all the actions, and displays the result.

…

• Hey, ChatGPT, what interesting happened in the world this week?
• Over the past week, several significant events have occurred in the world:
  – A $6 million bank robbery took place in Costa Rica.
  – An aide to Boris Johnson said he suggested a COVID injection to prove a point.
  – The robot killed the worker who was examining it.
  -The Prime Minister of Portugal resigned after the arrest of his chief of cabinet.
  -Four people are accused of stealing a gold toilet worth $6 million.

An example of using a search engine by a model. The fifth news actually appeared on the day the article was written - so the material is fresh!

The most curious readers might wonder: how does this actually work? How do you “connect” the real world to a language model that can do nothing but read and write text? To answer this question, we need to recall two facts that we discussed in the first article, “How ChatGPT Works“:

• Modern language models were trained to follow instructions.
  .
• Modern language models have a good understanding of programming concepts and can write code reasonably well. (Of course, they’ve read the entire internet! So many heated discussions on developer forums, and documentation has been helpful too, of course.)

Based on these facts, the following idea emerges: let’s write an instruction that shows the model how it can interact with, say, a calculator using code. The external program will simply “read” the model’s output in words and perform the corresponding actions.

…

You are a middleware designed to translate text commands into commands for a calculator (an external tool).
In order to give a command to the calculator, you must write: “Send to the calculator: (…)”, and in brackets indicate:
1. The first argument of the operation as a string.
2. The second argument of the operation as a string.
3. Mathematical operation as a string.
For example, “three by two” should become Pass to calculator: (“3”, “2”, “*”)

Write “OK” if you understand the command.

OK

For example, we instruct ChatGPT on the format of the response we expect. The only way for it to satisfy the user is to follow our instructions and do exactly what we asked (even if we presented the instructions in a peculiar order).

It sounds incredibly simple, but it works even for complex plugins! It may be hard to believe, but this logic is exactly how a browser is connected (when the text on the screen is translated into plain text, and the model decides where to “click”). For all the details about training the model to surf the web, you can read Igor’s article “ChatGPT as a Search Tool.“

Another one of the most useful and popular tools available to the model is the Wolfram Alpha math engine, familiar to every tech-savvy student (humanities folks, you can relax for now). Now, any complex calculations are no obstacle for LLM!

Research shows that GPT-4 can even handle the management of an automated chemical laboratory and carry out the synthesis of substances of varying utility, but that’s a different story.

…

GPT-4 was connected to a tube management system (depicted in the top left). It was given simple tasks described in natural language to construct specific shapes from reagents. The model successfully passed the tests.

The only problem with tools (plugins) is that the model can get lost if there are too many of them. It’s not always clear in what sequence to use them and which one to choose specifically. The model’s skill is more akin to “good” rather than “excellent.”

That’s why they’ve now been organized into different chats: in one, you can surf the internet, in another, you can program, and in a third, you can write a term paper with Wolfram (just don’t tell your professor what you’re up to). But over time, the model has improved, and now it can do everything at once, without compromises!

Autumn 2023: Text and image model Dall-E 3, or a quest to generate the perfect cheburek

A separate product that OpenAI recently introduced at the end of September is the generative neural network DALL-E 3. Like its first and second-generation predecessors, it generates images based on input prompts. However, most similar neural networks have a rigid limitation: the longer the prompt (input text query) and the more details it contains, the less the generated image corresponds to the description.

Therefore, prompts often consist of just 1-2 sentences (sometimes even a couple of words), and most of the details are left to the model’s interpretation: it will depict the object as it envisions it. While the tool can be useful for artists/designers, it doesn’t fully meet their needs, as it’s challenging to achieve something that entirely matches the artist’s vision and intended composition.

…

For example, take the painting “Théâtre D’opéra Spatial,” which won the Colorado State competition in 2022. The artwork outperformed others created by human artists, but it required over 600 prompts to the MidJourney model to bring it to life!

OpenAI has taken a huge leap forward here: now, DALL-E 3 understands giant prompts and creates images that precisely match the given text. Let’s take a look at an example from the product’s landing page:

…

Of course, the best example is selected for advertising on the official website, and such intricate generations may not happen every time. However, based on initial subjective tests and online reviews, the attention to detail by this new neural network is still impressive.

The reason DALL-E 3 is featured on this page—although it seems unrelated to ChatGPT and large language models—is rooted in the principle of its operation. DALL-E 3 was developed from the very beginning based on ChatGPT, as this language model generates detailed and effective prompts for DALL-E 3 (based on your “improvised” requests). Just briefly tell ChatGPT what you want to see, even in two words. It will rewrite the prompt, enrich it with details, and only then pass it to DALL-E 3. The integration works exactly like the “plugins” idea described earlier!

AI literally takes on part of the prompt engineering work, replacing the lazy human while also suggesting new ideas for images. You write “cheburek,” and you get (we apologize in advance to anyone who is hungry right now!)…

…

Generated prompt: “A freshly made cheburek on a wooden cutting board, half-cut to reveal the juicy meat filling inside. The dough is golden-brown and crispy, with steam rising from the filling. The background is a rustic kitchen setting…”

What’s more interesting is how this model was trained. We don’t have all the training details, but OpenAI shared the most crucial differences. As far as we know, this is the first time a model of this scale has been trained on synthetic data rather than human-created data.

You heard it right—95% of the image-text pairs (the data the model is trained on) were generated by GPT-4-Vision, announced in the spring. The model looked at images from the internet and wrote several long descriptions, repeating this process billions of times. That’s how models started helping train other models, and there will be no stops on the path to singularity!

Fall 2023: AI assistant from the world of science fiction

Remember Siri, that virtual assistant? Right after its debut, it seemed like we were on the brink of a world filled with super-smart and cool robo-assistants that understood us effortlessly and could do a thousand things. However, over the more than decade-long history of Apple’s product development, it feels like there haven’t been any mind-blowing updates. Siri, or “botina” as it might be called now, still fumbles, confusing a call to “my mom” with “my grandma”…

Meanwhile, in September, an update for the mobile app of ChatGPT was released, allowing it to see, hear, and speak. Now, the most powerful neural network of our time has convenient communication interfaces with you. And most importantly, it understands dozens of languages, can respond in them, and is capable of “Binging” under the hood.

Here’s an example where a guy on Twitter (oops, sorry, X) is trying to learn Russian—notice that the app responds to him in different languages without changing the voice. Overall, it looks really cool, and Jarvis from “Iron Man” is probably gnashing his hat in envy!

The technology behind this also powers image-related tasks. You can upload several photos (even documents), highlight an interesting part, and ask ChatGPT about it. How to fix a bicycle? Which key from the set should you use (to avoid getting scolded by your dad)?

Example with an image: you can outline a specific area in a photo and ask ChatGPT, “What’s going on here?!”

Some even asked for directions to the nearest store from a photo! No, it’s not like ChatGPT knows every street, but understanding urban planning and looking at signs, it could suggest how to get there.

The same idea is the basis for the product of the company Be My Eyes—it helps blind or visually impaired individuals with tasks related to vision, whether it’s finding keys or something more important. Volunteers used to work there, but now they are being replaced by GPT. In the near future, technology could literally become the eyes to the world for someone without the ability to see.

Here and now: GPT-4, turn on Turbo acceleration!

Well, here we are, it seems we’ve reached the present day. On November 7th, an event occurred that prompted us to write this piece—the OpenAI DevDay 2023 conference, where over a dozen small and significant updates were presented for almost every product of the company. As we’ve seen before, over the last six months, GPT-4 has significantly advanced, enriched with auxiliary tools and interfaces.

Some companies have already started implementing it in their businesses and even building separate products exclusively on this technology. However, it still has many limitations, and developers wondered—what specifically would be revealed at the long-awaited DevDay?

…

OpenAI started with a trump card: GPT-4-Turbo. Seven improvements were announced, but many of them have a technical nature (after all, it’s a conference for developers), so we’ll focus only on the most crucial and interesting ones.

If you’ve been using ChatGPT for a whole year, you might have noticed that it doesn’t respond (or hallucinates) to questions about events after September 2021. If you wanted to process such information, the Bing search mode came to the rescue. Alternatively, you could manually upload a document for the model to “read” and provide relevant responses.

During the conference, it was announced that the model’s knowledge has been updated all the way to April 2023, and they no longer plan to leave such significant temporal gaps in the model’s memory. This means that approximately every 1-3 months, the model’s knowledge will be brought up to a more recent moment. The key is to ensure that nothing from the past is forgotten in the process!

…

Case 1: When knowledge is limited to September 2021
Case 2: When knowledge is limited to April 2023

Rumor has it that when the poor model was forced to “learn” the news for 2022, terrible screams were heard from the OpenAI server room…

In addition to this, the model’s capability for file uploads has been enhanced. Now you can upload your files, totaling several gigabytes, to the OpenAI website. During response generation, the model will first search for relevant information on the uploaded files and then provide an answer. This doesn’t mean that the problem is completely solved for all types of questions, but it will certainly improve the quality of responses in domains of interest.

Furthermore, the model’s context length has been significantly increased to 128,000 tokens, equivalent to over 300 pages of text. Now you can engage in a sequential dialogue with ChatGPT for a couple of weeks, ensuring that the model won’t forget details discussed in the previous week.

It’s worth noting that this is currently the largest context available in the market from private companies. Prior to this, Anthropic with the Claude 2 model held the lead with a context of 100,000 tokens. However, among open (but somewhat less sophisticated) GPT models, “giants” with a window of 200,000 tokens appeared just yesterday.

…

A chart comparing different models before the GPT-4-Turbo burst onto the dance floor.

The reader may naturally wonder: what’s the point of having such long chats, and what’s the benefit of these extended interactions? Let’s explore a few scenarios:

1. Development Assistant with Project Understanding: In the prompt for a development assistant, you can input not just one file or a code snippet, but an entire project or a significant portion of it. In this case, the AI will have a better grasp of the project, understand which prompts to provide, anticipate potential bugs, and so on. A similar logic can be applied to a legal assistant reading, for example, all tax legislation in one go.
   .
2. Extensive Instruction Writing: Writing a massive instruction as long as a book, describing all the nuances of a given task. Often, the model might overlook a human-understood detail, and the prompt might lack space for nuances. With the increased context length, these can now be accommodated.
   .
3. Enhanced Few-Shot Prompting: One of the most popular and effective ways to improve the model’s response quality is few-shot prompting, where the model is shown a couple of dozen examples of what needs to be done before posing the task. It’s understandable that such a set can’t cover every block of logic, but if you expand it to thousands of examples, the situation might undergo a radical change.

…

Here’s how a few-shot example looks: the prompt contains three examples of sentiment analysis for reviews (2 positive and 1 negative). In this case, ChatGPT predicts an incorrect answer. Perhaps loading not just 3 but 3000 examples into the prompt would be beneficial in this context.

In summary, the primary goal of such changes is to enhance the overall quality of ChatGPT responses through more detailed task descriptions, whether they be examples, instructions, or comprehensive work context.

Let’s make a careful assumption that those who predicted the imminent death of prompt engineering before models with lengthy context emerged likely just lacked imagination. In essence, we haven’t really started to fully write (and automatically generate) prompts!

By the way, Sam Altman emphasized that the model is smarter than the regular GPT-4. It’s already available in the official UI at chat.openai.com, so give it a try and share your impressions: has it improved or not?

API access to all models and price reduction: Christmas gifts for developers

Just as great power comes with great responsibility, a large prompt comes with a hefty bill for using GPT. Paying for API usage (the interface developers use to access GPT) depends on both the length of the prompt and the generated text. This is quite logical since it directly impacts the amount of computation required for the neural network to function.

…

Sent a request to the GPT API for 128 thousand tokens:
Case 1: Before
Case 2: After

That’s why the announcement of a price reduction for the Turbo model received the most applause at the conference. Using such a model is now three times cheaper for the text from the prompt and twice as cheap for the generated tokens (usually fewer). Why is such a distinction important? As mentioned earlier, sometimes you want to cram a lot of details and examples into the prompt. Now, for the same price, you can fit three times more, and it should work better, or you can simply save on usage costs. Either way, it’s all positives!

In addition, developers now have access to the API for all the mentioned models: for working with images (GPT-4-Vision), generating images in Dall-E 3, and generating voice from text (with voice-to-text translation already available earlier, now improved with a new model). The API is a way for an ordinary person to access closed models running on some server and get results. So now, every developer can integrate these technologies into their application in parts.

…

Here’s an example of how GPT-4-Vision helps with home inventory by identifying IKEA items. However, it made one mistake — attentive readers are invited to find the inaccuracy! You can read the full story here.

DIY enthusiasts have already created several interesting prototypes in the first day. For instance, an AI soccer commentator! Two frames per second are taken from a video recording, hundreds of extracted frames are fed into GPT-4, and it, in turn, writes the commentary as if spoken by a commentator. Then it is voiced by one of the six OpenAI voices, and here’s the result.

It’s not as emotional as a Spanish commentator, but it’s only 2023, give AI a little discount and some time! Especially since workers in the voiceover industry are already complaining that their jobs are being taken away.

The idea is so straightforward that almost simultaneously, a second cardboard commentator appeared. This time, for the popular online game League of Legends. The quality of the generated speech is higher, and the comments are relevant to the game strategy.

And a few more examples of witty pranks: an app to evaluate the correctness of yoga poses, a browser window Q&A (or any other application), a chat with video on YouTube or even with your webcam, creating and animating a GIF (try it yourself here), and a favorite: criticizing a website’s design (when creating this bot, we hope no Teima Lebedev suffered). Of course, the cult and highly useful hot dog / not hot dog classifier from the TV series “Silicon Valley” was also created right away.

Yes, it’s not something that blows the imagination, and similar apps on phones have long existed. However, what’s important here is that it’s all a mix of two or three different models, connected in one line of code. Now these tools are available to everyone, they work on a wide range of tasks (often even better than specialized systems designed to solve one specific task — for example, finding cats and dogs in a video), and you can whip up a prototype in an hour. At the same time, the technology becomes more and more accessible.

On Twitter, even a meme started circulating, mocking startups that were thinly veiled as minimal value-adds compared to OpenAI products.

…

The picture was made in Photoshop, but it’s still lol: you can’t argue with that.

For example, sites like ChatWithPDF / AskPDF allowed users to upload a file (even a large one, up to 100 pages), and then ask questions about the document, with the answer generated based on the provided source. Too lazy to read a 50-page report on work? Study it in 3 minutes! However, the technology was very basic — with some effort, you could whip up similar functionality in an evening.

OpenAI scratched their heads and said, let’s give every user the ability to chat with documents? Boom, and the small knee-joint startup evaporates, as if with a snap of the fingers. However, real startups developing domain expertise and providing greater value without auxiliary technology are not threatened by such a fate… well, at least not yet, lol.

Support in legal cases regarding copyright, or how to use the fruits of neuron safely

We live in a time when it’s sometimes challenging to distinguish true art from gimmicks. Although debates on this matter have been ongoing for at least a century (since the appearance of Malevich’s “Black Square”), in the era of AI, these debates are particularly acute.

…

While disputes about the legality of using text and images from the internet for training neural networks are ongoing in major jurisdictions, large companies see risks in their use. What if a copyright infringement lawsuit arrives tomorrow? What if the generated image for a magazine cover or movie poster is not truly original?

Understanding and sharing the concerns of businesses, key technology providers are moving to address them.

For example, if a third party sues a commercial customer of Github Copilot (roughly speaking, it’s ChatGPT for programmers) for copyright infringement due to the use of the product or its results, Microsoft will defend the customer in court and, if necessary, pay fines or damages. Similar announcements have been made by Adobe for the use of generative functions in Photoshop (the Firefly model), Google for almost all of its products, IBM, and others.

…

Case 1: Pathetic parody
Case 2: Unique original

In general, now, if someone accuses you of inappropriately copying other people’s ideas, feel free to answer, “Google allowed me to do all this!”

And at the DevDay conference, it was announced that OpenAI is also entering this game by launching the Copyright Shield program. Unfortunately, it doesn’t apply to all users, only Enterprise and developers. In other words, if you generate something on the official website, it won’t be covered by protection unless your company has a separate partnership agreement with OpenAI.

Interestingly, just a couple of weeks before the announcement, three artists filed a lawsuit against technology companies (Midjourney, Stability AI, and DeviantArt) accusing them of copyright infringement. In turn, these companies filed a motion to dismiss the case, and a U.S. District Court judge granted this motion.

The main reason for this decision is that the artists did not register copyright for each of their works. However, the court also provided recommendations for adjusting the claims. What happens next will be revealed in the upcoming episodes!

…

By the way, if you’re concerned about your data, Sam Altman assured that OpenAI does not train models on user data. This is true by default for businesses and developers using the API, while regular users need to uncheck a specific box in the ChatGPT website settings.

The mission of Microsoft and OpenAI: happiness for everyone, and let no one leave offended (or something like that)

And now, the most significant part of the presentation featured Satya Nadella, the CEO of Microsoft. Together with Sam Altman, they discussed the partnership between the two companies and their shared vision. Microsoft’s official mission is to “empower every person and every organization on the planet to achieve more.”

The development of tools that enhance work efficiency and expand capabilities aligns perfectly with this mission. Intelligent AI assistants based on ChatGPT are already contributing to this goal, as evidenced by research studies from MIT and Harvard University. So, what’s next? What is OpenAI’s plan?

Globally, their vision involves creating AGI (Artificial General Intelligence), a universal artificial intelligence that benefits all of humanity. Before you start imagining Terminator scenarios, let’s clarify. AGI has many definitions, so it’s crucial to set expectations correctly.

OpenAI’s definition can be roughly summarized as follows: AGI refers to highly autonomous systems that outperform humans in most economically valuable work. Not so scary now, right? No Terminators (at least, not yet).

…

Satya and Sam look at you like you’re a bag of leather whose life they’re about to improve beyond measure with their highly autonomous super-smart AIs.

In this definition, there are several key components. The first is the autonomy of systems. They should operate with minimal human involvement, receiving a high-level formulated task. It operates on the “give the task and forget” principle. The second is a focus on the economic aspect, on increasing the efficiency of intellectual work.

The ultimate goal is to make it possible to simply tell the computer what final result you want to achieve, and it will independently come up with and implement all the necessary subtasks to achieve that goal. Systems with such capabilities in the field of AI are often referred to as “agents.”

…

The emergence and implementation of such technology will require extensive thoughtful discussions throughout society—what to do with people who lose their jobs? How will politics change? What rights will AI “workers” have? But for now, this is a somewhat more distant and uncertain future, and we are here and now. OpenAI, as part of the conference, talked about the first small step toward this future: GPTs.

GPTs: A glimpse of the next generation of AI agents

GPTs are customized versions of ChatGPT tailored for specific purposes. They differ from the original in three aspects: instruction, expanded knowledge, and available actions. You can program your GPT by simply conversing with it using natural language. This significantly lowers the entry barrier, as there is no need to deal with model training, integration of external tools, and so on—everything is ready for use. Let’s go through each aspect.

Instruction: It defines the “personality” of ChatGPT, what function the neural network will have, and what rules it will try to follow. You can either write your own prompt or leave it up to GPT based on your brief description.

When creating a bot, you’ll be asked what this AI should do. Sometimes, if complex logic is implied, the bot might ask more than three questions to clarify the desired behavior—even if you haven’t thought of everything yourself. And each time, the questions will be unique to your mini version of GPT.

…

Creating a new bot agent live: look at Altman’s sarcastic face, what is he up to?

Moreover, Sam, the former president of the prestigious Y Combinator accelerator, who has given dozens of lectures on business, often receives questions from startup founders. Now, he wants to automate his responses and instructs the bot to brainstorm the user’s business ideas, provide advice, and then conduct a roast on “why your business isn’t growing faster.” The GPT agent then rephrases this instruction, expanding it to 5 lines, specifying the style of responses and behaviour.

Next comes the “expanded knowledge” block of the model. Using the file upload button in the ChatGPT demo, a summary of Y Combinator lectures is uploaded. Now, all the information from there is available in text form when answering questions.

…

When answering a question, the model can now peek into the lecture notes and provide an answer based on the material. It’s like a student with a crib sheet!

In this way, in just 4 minutes, Altman’s major headache was resolved – now he can simply share a link to this bot with all the startups, and they won’t bother him with the same questions (although there is a suspicion that all these folks would like to get answers specifically from Sam, not from a neural network…). Any business can do the same, automating a good portion of customer support or even onboarding new employees.

The third component – actions – was not demonstrated in this demo, but essentially, it’s just an evolution of the plugin connection interface we discussed at the very beginning. You can write code implementing any complex logic and describe the models in simple human language when you want to use it.

The model, in turn, will make decisions on its own. This was demonstrated within the context of a travel assistant chatbot. The host uploaded a PDF file with tickets, GPT recognized it, and invoked a specific method for a website that displays the information on the screen.

…

The beige block above the map appeared only after the file was loaded. ChatGPT subtracted all the values in it from the PDF file itself, and then sent it to the server.

To a human developer, it might have been necessary to come up with some workarounds to answer the question, “How will I know if these are tickets, not, for example, a hotel reservation?” A GPT neural network, in this case, essentially removes the barrier of interpreting human-written text and acts as a binding agent, translating ambiguous and complex natural language into specific commands. The task of writing these commands for your website or product is (for now) still in the hands of programmers.

Towards the end of the presentation, the host verbally addressed the AI assistant and ordered it to provide $500 in credits for the use of OpenAI products to each developer at the conference (which, understandably, sparked enthusiasm in the audience).

…

GPT understood the command and, under the hood, did the following:

1. Called the function to retrieve all participants from the list registered for the event.
   .
2. Iterated through each participant, calling the method to credit their account.

So, there wasn’t any magic happening—both the function to retrieve the list of participants and the function to credit an account for each participant were written by a human (although a machine could probably do it too).

However, how to use them, when to use them, and how to combine them are decisions made by the AI based on the context of the conversation. So, instead of having two such functions, you could plug in a thousand, and ChatGPT would immediately start managing everything around. And here you thought, why do we need smart sockets and light bulbs?

Is OpenAI a future giant with an Apple-like ecosystem?

And right after that, Sam announced that at the end of November, the GPTs online store will be launched, where everyone, after passing moderation, will be able to share their creation. That’s why some refer to this announcement as an “iPhone moment” for AI applications (meaning an event that has the potential to become a turning point for the development of the entire industry).

…

This is what a beautiful store of the future looks like!

In the store, there will be top lists and a section of recommended GPTs—just like in the App Store. Ideally, specialized agents should “live” here. One teaches English, another helps with math for children, a third explains and vocalizes cooking recipes, and a fourth optimizes website SEO.

It will be very interesting to see which solutions will top the charts right from the start—will they be remakes of popular apps for Android and iOS? Or something radically new with AI specificity? We’ll be watching and keeping you updated!

A recent example is a GPT that writes an adventure story for you, where at each stage, you determine what happens next. And nothing in the story is predetermined! Like text-based quests from the 80s-90s but much more advanced. Additionally, illustrations for a segment of the story are drawn by Dall-E 3 directly in the browser to stimulate the reader’s imagination.

…

The prompt (on the left) with instructions for the bot has become longer. Additionally, a set of rules for text-based role-playing adventures (DnD) has been loaded into GPT. On the right, the model generates part of the story and then provides choices for further development.

One can come up with anything! Specifically, what attracts us, the authors of this article, the most are the possibilities of applying AI in education. Throughout the past year, teachers have been trying to combat cheating, especially in essay and thesis writing, especially since there are still no reliable methods for detecting text generated by a neural network. But what if we take the same tool and instruct it not to write an essay from scratch but to critique and provide advice on what’s already written?

Anyone can upload a file with their composition and receive a thesis list of “growth points.” Personalized feedback, with the machine acting as a teacher. While this might not help those who are simply lazy or unwilling to spend time, it can motivate people genuinely trying to improve their writing and receive a fair evaluation, pushing them towards new achievements.

…

The power of the technology here is quite evident. A teacher can create their own GPT for each class and topic. Some of them might even be interactive simulations in which students can immerse themselves; others may serve as tutors or mentors; and some may even act as team “partners” suggesting ideas.

For the best applications, OpenAI promises to pay developers. However, the monetization system remains unclear: access to GPTs (for now) is free for all ChatGPT Plus subscribers ($20 per month). Embedding something unique that cannot be copied into the bots themselves is challenging because they are language models that can still be easily deceived.

Someone could claim to be a super-secret OpenAI developer and request access to the bot’s internals (its prompt). Any prepayment request can also be circumvented by convincing the neural network that you have already paid; it’s just that it can’t get confirmation, but that’s not your problem. Let’s carefully assume that the main feature and uniqueness of the bots will lie in the plug-in tools (the functions that developers write), which external users cannot copy.

…

A jailbreak master class by Igor: managed to persuade the model on the second attempt, well, you get the idea. =) P.S. No cats or LLMs were harmed during the creation of this screenshot!

With the release of GPTs, OpenAI makes it clear that this is just the beginning. By adding actions to the bot, GPT can easily be integrated with other systems, such as email, messaging apps, or any website.

As a result, we might witness the birth of true agents that can interact relatively broadly with the world. However, both short-term and more distant risks are easily noticeable. If, in the near future, AIs are connected to an increasing number of systems, and we gradually trust them with more and more tasks, then… Well, let’s leave that for another time.

Translation: Epilogue: What the Coming Day Holds for Us

However, it must be acknowledged that the functionality of GPTs is currently limited by the capabilities of ChatGPT. The model has its limits, and occasionally, if not frequently, it makes mistakes, looks in the wrong direction, or writes something incorrect. On the other hand, users have become accustomed to this, and they are likely willing to give the neural network a second chance if it happens to make a mistake.

…

An important point to understand is that as soon as GPT-4.5 or GPT-5 is released with the same interface as GPT-4 (which serves as the basis for these GPTs-agents), all the applications already created will instantly (and almost certainly without additional costs) migrate to the new “engine.” The fact of migrating to a new, more powerful and capable base model will significantly enhance these applications.

Imagine that when you update iOS on your iPhone, not only does the browser start working 3% faster, but also your phone and the installed applications suddenly gain entirely new features automatically (and this is even without changing the hardware itself!).

The same kind of upgrade can be implemented here; and such a transition is logically expected in GPT—after all, OpenAI itself aims to improve agents, enhance their skills (memory, accuracy in choosing tools, reasoning, and so on), and in this sense, their goal aligns with the desires of developers. Sooner or later, one GPT will be able to call another, specialized one, and delegate a specific task to it… thus creating chains of agents.

…

I see these same chains of agents exactly like that!

It is quite possible that by 2025 or somewhere around that time, we will see much more advanced agents that, in some sense, will be indistinguishable from humans. Sam Altman even envisions AI being hired as a “remote worker” whom you will never see in person, but simply assign tasks to.

Of course, you’d pay them at the end of the month. Such a future might be in store for us, or maybe not—who knows? It’s possible that countries showing interest in AI regulations (at least the United States and G7 countries) might impose moratoriums on further technology development without the oversight of the “Big Brother.” Leading research labs might go underground and start operating from autonomous maritime data centers in neutral waters.

…

And it’s not even a joke: Del Complex has already presented a concept for a floating AI dreadnought, highlighting the ability to operate in an unregulated zone as the first item in the list below.

In short, guys, what do you think – is it already cyberpunk or not yet?

That’s it, thank you all for your attention! As usual, we look forward to your comments. If you don’t want to miss our next materials on the topic, we invite you to subscribe to the Telegram channels of the authors: Igor Kotenkov’s Syoloshny’s channel (for those who want to dive into technology) and Pavel Komarovskiy’s RationalAnswer channel (for those who prefer a rational approach to life but like it a bit simpler).

By Pavel Komarovsky
I write interesting things about finance at t.me/RationalAnswer
The text is posted with the permission of the author.
The original material is here.