Author: Shabbir Chunawalla

In India and elsewhere, there is a growing non-music audio market consisting of audiobooks, podcasts and audio series. Here two startups Pocket FM and KuKu FM are in audio entertainment segment.

Pocket FM offers bite-sized episodes of audio stories and novels in regional languages. It wants to deepen push into the US and expand into Europe and LATAM market in 2024. It follows a freemium model where the revenues are generated through both subscriptions and advertising. There are micro-transactions — a user can unlock a chapter of a novel for as low as Rs.9. In 2023, there were 20 million transactions and 75 billion of minutes of streaming worldwide.

KuKu FM largely focuses on audiobooks and podcasts in regional languages. It has signed an exclusive contract licensing deal with Storytel (Stockholm-based). It will offer translations of English books in regional languages.

Mustafa Suleyman, as we know, was the co-founder of AI research lab DeepMind. DeepMind since then has been acquired by Google in 2014. He continued to work in DeepMind until 2022. It wanted to create AI that will not veer into racist, sexist or violent behaviour. He left DeepMind in 2022 to cofound Inflection AI, an ML and generative AI company. Reid Hoffman (Greylock’s) too was a co-founder of Inflection AI. In 2023, Inflection AI launched a chatbot named ‘Pi’.

Microsoft has now appointed Suleyman (2024) as EVP and CEO of its newly created consumer AI unit, Microsoft AI. Several members of Inflection AI’s team have been appointed to the division, including co-founder Karen Simonyan.

He has co-written a book, The Coming Wave, that examines AI’s promise and the need to limit its potential perils.

The companies in AI space, namely Google, Microsoft and Apple are forming alliances so as to capitalize on generative AI. The idea is to generate revenue by marketing suitable consumer products and capture the market share. Each company does not have all the ingredients which can be assembled together to capitalize on generative AI. The ingredients are computing power, top-of-the-line AI models, trustworthy products and ways of getting them to people. The companies continue to search worldwide for talent and promising startups.

Google’s products have serious errors and biases. Microsoft has not been skilled at building exciting consumer products apart from video games. Apple is years behind in AI.

Big Tech is incapable of innovating the entire generative AI ecosystem single handedly.

Microsoft has infused AI into Bing search engine, Windows, Office and other products such as digital assistants under the Copilot brand.

This is the reason why Suleyman has been roped in. They went to craft in a true end-to-end product experience. Suleyman compares this to sculpting — recognizing when a piece of technology is ready and how to dress the experience that it becomes accessible and trusted.

Is Suleyman the right person to do this? He has developed a chatbot called Pi, which attracted a million active users. However, his startup Inflection never found a business case.

Bitcoin prices reached a high $73000 mark. The market anticipates a high demand, while the supply is limited. It is all because of the halving event which occurs once every four years, The event is steeped into the design and philosophy of scarcity. Satoshi Nakamoto, the creator of Bitcoin, designed the crypto to have a finite supply of 21 million coins. The issuance of new coins is reduced. Halving is significant as it cuts by half the rewards the miners receive while validating the transactions. It reduces the rate at which Bitcoins are created, suppressing the supply of new Bitcoins. In 2009, miners received 50 Bitcoins for every 10 minutes. There were three halving. They now receive 6.25 Bitcoins every 10 minutes.

Halving occurs after mining 2,10,000 blocks. Mining is the process of making computer hardware do mathematical calculations for Bitcoin network to confirm transactions and increase security. In April 24. there will be next halving. The block reward will be restricted to 3.12 Bitcoins for every 10 minutes of mining.

Bitcoin pricing, among other things, is scarcity driven. By the year end (2024), the prices may hit $150,000 and $250,000 in 2025.

As value increases, despite reduction in rewards, profitability of miners may not get affected. Miners may have to reduce power consumption and may have to improve hardware efficiency.

UBS, the bank which took over the collapsed Credit Suisse, has a balance sheet worth $1.695 trillion. It overshadows the Switzerland’s home economy. It is twice the size of the GDP of home country. HSBC’s balance sheet in terms of leverage of exposure is worth 83 per cent of UK’s GDP. BNP Paribas is equal to 72 per cent of France’s GDP. At the same time, JPMorgan Chase, the world’s largest bank is just 17 per cent of US GDP.

UBS should not be in distress, as it would burden Switzerland unbearably. There are three other systematically important Swiss banks — their sizes range from 15 per cent to 37 per cent of GDP.

Credit Suisse failed because of its top management (board and executives). There were years of bad strategy and faulty management decisions. The executives kept changing but there were new scandals and losses with each new batch of executives. Far too much capital, and far too less return dragged it down. Discipline and risk-return were ignored.

The regulator conducted several investigations and issued several reprimands. There were a few criminal charges. There was action against the staff. As the whole banking sector would be affected adversely, regulators do not name and shame people and institutions. There are punitive fines and dressing down in regulator’s offices.

The gradual collapse could be seen coming, but regulator cannot put restrictions. The intervention comes when the institution is on the brink of collapse. Despite being aware about the shortcomings, the regulator could not direct the bank on the right path. The management refused to face the truth.

There was loss of liquidity at Credit Suisse. Depositors fled. There were collaterals to raise funding. It was not possible to transfer the collaterals from the group holding company legally to its overseas units.

The US, UK and Europe have realized this issue and have asked the banks there to use liquidity facilities to prevent the speedy runs on banks in modern times.

The chip making company Nvidia is now the third most valuable company in the US, behind only Microsoft and Apple. This Santa Clara-based chip maker has earned the title of the world’s most valuable chip maker, eclipsing the celebrated competitors such as Intel and AMD.

There is an AI boom and there is edge computing. The firms are moving from exploration to deployment. AI computing basically requires high performance graphical processing units (GPUs). Traditionally computers used central processing units (CPUs). Intel and AMD dominated the CPU market. GPUs are relatively new additions to the computer hardware market. These were initially sold as cards that were plugged into a personal computer to add computing power to an Intel or AMD CPU.

The graphics chips can handle the kind of surge in computing power that is needed in high-end graphics for gaming or animation applications. Standard processors cannot handle this surge. AI applications too demand high computing power. In their backend hardware, these apps are becoming GPU-heavy.

In most advanced systems for training generative AI models, for every one CPU, at least half a dozen GPUs are deployed. The equation when GPUs were just an addition to CPU has completely changed. This lead will be maintained by the GPUs in the near future.

Nvidia first popularised the term GPU in 1999. Its chip was called GeForce256. This chip was coveted for graphics. These chips were more expensive than most CPUS (on a per unit basis). It resulted into better margins. TSMC, the Taiwan-based foundry specialist is the important player in the backend semiconductor business. Intel, AMD, Samsung and Qualcomm are the front-end players.

The most popular AI chip of Nvidia is H100, which was launched in 2023. It has 80 billion transistors. The company has now introduced B200 Blackwell, the new chip. It has 208 billion transistors. It can do some computational tasks 30 times faster than the current blockbuster H100. The new chip with its more computational power and optimised power consumption will strengthen the dominance of Nvidia in the niche space. It is twice as powerful while training AI models, and five times more capable while inferencing. (the inferencing is done by models such as Gemini, or ChatGPT while tackling queries and generating response).

A training GPT model (which powered ChatGPT) had 1.8 trillion parameters and 8000 Hopper GPUs. It consumed 15 MW of electricity. The new 2000 Blackwells can do the job while consuming just four MW of power.

Major buyers such as Google, Amazon, Microsoft and OpenAI are expected to use the new chip in their cloud-computing service and in their AI products.

Nvidia is ahead in AI race because of its hardware as well as its proprietary software that facilities the leverage of its GPU hardware for AI apps. Nvidia also has developed systems that back its processors and software to run all of this. It is thus a full-stack solution company.

Apple and Google are reported to be in talks for empowering Apple’s iPhones with license to use its LLM Gemini, a generative AI model. Apple has reported to have discussions even with OpenAI for using its model on its iPhone.

Since long, Google was being paid billions of dollars by Apple per year to make Google Search a default option on Apple phones and other devices. If the present Gemini deal comes to fruition, Apple will have to pay Google substantial fees per year. Though Apple itself is fiddling with AI research (Ajax), the technology remains far inferior to the tools developed by Google and OpenAI.

The terms of the agreement have not been finalized yet.

Gemini will be on firm foothold if the deal materializes, as it will have access to more than 2 billion devices in active use. Samsung has already rolled out new smartphones infused with AI features powered by Gemini.

Yotta Data Services is in the data centre business since 2019. It was co-founded by Sunil Gupta, the CEO. The datacentres are being converted into compute units to produce intelligence. In past, water on one side generated electricity on the other side. These days we put data on one side to get intelligence on the other side. India slowly could emerge as the AI factory of the world by following this strategy. Yotta has the backing of Niranjan Hiranandani, the real estate tycoon. Yotta provides access to organisations such as Wells Fargo to data storage and computing power so that they can scale up or down without creating their own infrastructure.

To convert data centres into AI factories, what you require are AI chips from Nvidia. These are used to train the LLMs and build applications such as ChatGPT (OpenAI) and GitHub Copilot, the coding assistant of Microsoft.

Yotta wants to allow the least expensive access to Nvidia AI chips to Indian firms and startups. The smaller companies with tight budgets can get access to AI chips by giving equity, rather than cash.

The global AI market currently at $160 billion plus is likely to reach $2 trillion by 2032. Thus, for AI these are early days, and there would be a gold rush.

Yotta has already received an eagerly awaited consignment of 4000 H100 chips from Nvidia. Each GPU costs $30,000 to $40,000. Each unit looks beefy and bulky. The arrival of the maiden consigment is a dream moment for Yotta. It has planned to receive 16000 more chips by June 2024, so as to have 20000 such chips. This nowhere compares to huge buys of Facebook — 3.5 lac H100s by the end of 2024, and Microsoft buys of tens of thousands.

India is receiving special attention after a meet between Nvidia CEO and our PM. The supplier realizes that India is a big market, with data and talent, and will receive priority treatment.

Yotta’s consignments kept the custom officer’s busy because of its high value. There was additional paperwork and so many bureaucratic approvals. On arrival at Yotta’s Mumbai suburban office the consignment boxes were adorned with flowers and vermilion and hymns were chanted by a priest.

AI research in India will receive a boost by having infrastructure which is supercritical.

A well-known computer scientist Ray Kurzweil predicts the arrival of AGI by 2029. He considers this to be a conservative estimate. It may come next year and the year after. Kurzweil was the one who predicted in 1999 that AI will match human intelligence by 2029. In all these years, people called it a crazy guess. Stanford held even a conference to deliberate on this. Some people made it a distant goal that may take a century.

Musk agrees with Kuzweil and is of the opinion that AI will be smarter than human beings put together by 2029.

People always underestimate the pace of growth of technology. It was said technology doubles in fourteen years, putting the growth at 2 per cent per year. Compute speed has increased tremendously — 35 billion calculations per second. It was just 0.00007 calculations per second in 1939. It is a 24 quadrillion fold increase.

The company behind Worldcoin’s founding member Saturnin Pugnet says that Fusion and AGI are happening by 2030.

When Q was leaked by OpenAI, everybody thought it is AGI. OpenAI is keen to build AGI and considers it as one of its core values.

It all depends how AGI is defined. The evaluation parameters are still evolving. The definition shifts over a period of time. The picture looks hazy.

Devin, an autonomous coding agent, introduced Cognition AI which completes the tasks. It autocompletes tasks and writes apps within no time. It is a massive breakthrough.

Laama3 may have AGI. The company, Facebook, has thrown its hat in the ring.

LeCun has said that it is not as simple as turning on a machine to have AGI. It is going to be a gradual process.

The Artificial Intelligence Act has been approved by the European parliament channelizing AI in human-centric direction where humans are in control of the technology. The idea is to leverage the technology to unlock human potential.

Big tech companies too are in favour of regulation of AI and at the same time the rules framed must not hamper the startups..

There are other European regulations to protect consumers, and AI Act too would forward the same concept taking a risk-based approach to products infused with AI.

The vast majority of AI systems are expected to be low risk (content recommendation or spam filters).

Companies can choose to follow voluntary requirements and codes of conduct.

High-risk AI systems (such as medical devices or critical infra-structure) have to meet requirements of using high-quality data and providing clear instructions to users.

Some AI users are prohibited since they pose an unacceptable risk (social scoring system, predictive policing, emotion recognition system). Facial recognition using AI-powered remote biometric identification is also unacceptable except for security uses and serious crimes.

While using generative AI, there should be detailed summary of media and data used. There should be labelling of AI-generated deepfake pictures, video or audio. The French and German governments pushed back against some of the strictest ideas for regulating generative AI, since this will hurt European startups (Mistral, Aleph Alpha).

There should be extra scrutiny for the powerful and large AI models carrying systemic risks.

European Union approves the world’s most comprehensive AI rules. There is absence of any legislation from the US, and hence this Act could set the tone for how AI is governed in the Western world. Europe is a trend- seller in trustworthy AI.

Europe has limited digital tech industry, and relatively low investments compared with Big Tech from the US and China. There is considerable work ahead. European Union is in the process of setting its AI Office.

The new law seeks to ensure the Charter of Fundamental Rights of the European Union, 2000 and other European laws govern the provision and use of AI within the EU.

The Act applies to AI providers of service in the EU or in a third country. The footprint of use in EU is necessary. Article 5 of the Act prohibits an AI system that subliminally affects the person’s consciousness or deceptively manipulates behaviour affecting the informed decision making, so as to cause harm. This is important in this age of fake news, targeted algorithms and powerful social media. All this affects consciousness of human beings, and makes them act which they would not have otherwise.

A couple of decade’s AI-enhanced growth in the world will make the world a changed place. AI is the most transformative innovation, and hence disruptive too. Still, all the exuberance over AI is still premature. We can get swept away by the excitement it has generated and the intellectual achievement that accompanies it.

AI’s full implications are yet to be understood. The idea is to make AI models that express as well as the humans. All this works well for marketing AI. There are two issues — AI as a tool that facilitates decision making and AI as a decision maker itself. These two are worlds apart.

It is to be seen how far it proves itself to be a good decision-maker, but the problem is that AI models hallucinate. Even those who design these models do not understand why they act stupid.

AI cannot make factual judgements. It all is a matter of values. AI handles complications by attaching values to actions and/or consequences. However, the model infers these from consensus (based on the information it is trained on). Alternatively, it infers on the instructions issued by its users/designers. Both these do not have any ethical authority.

AI’s arrival is ill-timed, as there is no distinction these days between facts and values. It is difficult to define objectivity.

These ideas corrode what AI claims to know. There is further push by designers to achieve cultural realignment.

AI models, so the companies think, must be socially beneficial, despite their reasoning power or instructions. A model has to choose between what is true and what is socially beneficial. AI’s ‘truth’ must be ‘gospel truth’ since it is smart. However, models such as Gemini do not follow this maxim.