Author: Shabbir Chunawalla

Google has introduced its new quantum computer chip called Willow. As we know, quantum computing relies on the laws of quantum mechanics. It handles data differently. Instead of bits (used in classic computers), it uses quantum bits or qubits. Qubits can exist in multiples states at the same time (such as 1 and 0, and anything in between). Qubits are not switches or transistors. They are actually elements that have quantum mechanical behaviour. Willow has 105 qubits and advanced error correction.

Quantum computing is amazingly fast. Google’s Sycamore quantum processor could complete a mathematical equation in three minutes, which would have taken 10,000 years on a supercomputer.

Google fabricated its previous quantum chips in a shared facility at the University of California (Santa Barbara). The current chip Willow has been built at Google’s own facility.

The problem with qubits is the rapid exchange of information with the environment. Thus, it is difficult to protect information needed to complete a computation. Thus, the more qubits one uses, the more errors will occur. Researchers are working on how to reduce the errors while using more qubits for the last 30 years.

Thus, errors are the greatest challenges in quantum computing. The arrays of physical qubits that have been tested are 3×3, 5×5 or 7×7. Each time they used the latest advances in quantum error correction, they reduce the error rate to half. In other words, they have achieved an exponential reduction in error rate.

Quantum technology has the potential to increase computational power exponentially. It gives more accurate predictions. It provides insights to transform communication networks. It facilitates optimizations. AI models can be trained with fewer data points. It will facilitate collection of AI training data. It can also solve data puzzles at the heart of encryption protection. It will thus make all our systems and data vulnerable — a big blow to cyber security and cryptocurrency.

Willow is a major breakthrough. It could pave the way for a large-scale quantum computer. Still at present, it is an experimental device. The whole field of quantum computing is experimental. It makes use of the field of particle physics to build a powerful computer.

The silver lining is that Willow is able to drive errors down while scaling up the qubits. It will lead to further progress in this field. At present, we are closer to running commercially relevant algorithms that cannot be replicated on conventional computers.

Apart from Google, IBM, Microsoft and Amazon are working on quantum computing systems.

Experts believe quantum computing will make encryption obsolete. However, Google has declared that the Willow chip is not capable of breaking modern cryptography. In future, a ‘cryptanalytically’ relevant quantum computer could jeopardize civilian and military communications, undermine control systems for critical infrastructure and defeat protocols for most internet-based financial transactions. But Willow is not CRQC (cryptanalytically relevant quantum computer).

OpenAI is at it, building AGI, one frame at a time. The turning point happens to be Sora, where video is leveraged to grasp the real-world logic and dynamics. It is a calculated attempt in the direction of AGI. It provides a learning environment. Sam Altman has dedicated ’12 Days of OpenAI’ and this is what he expressed on the third day.

Text-based prompts have limitations. Video stimulates AI systems to become more responsive to the environment. The early version of Sora is likely to commit mistakes, but it has the potential to advance AI capabilities and augment creativity.

Sora’s current version uses storyboard-based sequencing. It moves AI beyond generating videos. It should understand and interact with complex environments. Yann LeCun from Facebook is not optimistic about video generation since it does not understand the physical world. It is just one more video, rather than causal prediction. It should involve abstract representations of continuations in a scenario. Facebook’s architecture is joint embedding predictive architecture (JEPA). It focuses on predictions in representation space, rather than pixel reconstruction. Pixel reconstruction offers better performance in downstream tasks.

Sora could be a game-changer in the pursuit of AGI. It leverages video as a medium for training models in real-world logic and dynamics. The ultimate goal is to develop AI that understands and interacts with the world as humans do — one frame at a time.

Sora paves the way for AI systems that can reason and interact with both virtual and real-world settings.

Maybe by the end of ’12 Days’ campaign, Altman announces a breakthrough in AGI. That endorses Altman’s previous predictions of reaching AGI as soon as 2025.

Competitors too are working on this. The race is exciting. It is a race for reinventing intelligence. It is going to extremely consequential. It is going to hard, but not impossible.

It is necessary to be aware of the published research work in international research journals to move ahead in any area of knowledge and to improve the quality of R&D being undertaken in the country.

The international journals are costly. All of these cannot be accessed by the researchers, students and faculty members. The central government has taken ONOS initiative — One Nation One Subscription. It covers 6300 government higher educational institutions (HEI) and R&D institutions. This initiative gives access to 13000 scholarly electronic journals from 30 publishers across the world on one platform (which will be active from January 1, 2025). It covers all reputed publishers. One rate has been negotiated with them. The government has allocated Rs. 6000 crores for the calendar years 2025, 2026 and 2027. The implementing agency is INFLIBNET in Gandhinagar (Information and Library Network Centre).

To avail of this facility, the centrally funded institution must be registered on Anusandhan National Research Foundation (ANRF). The scheme will benefit 18 million individuals within the academic community.

ONOS will also provide a platform to Indian researchers to publish their research articles in these journals without article processing charges (APC).

ONOS provides an opportunity to get the articles published in renowned journals without paying any charges. In the open access system, the articles are published by paying huge amounts for processing. It is a shot in the arm for research in India.

There is a well-known test for AGI created by Francois Chollet in 2019. It is called ARC-AGI, short for Abstract and Reasoning Corpus for AGI. It is designed to assess whether an AI system can efficiently acquire new skills outside the data it was trained on. It shows how close we are to general intelligence. Till date, the best performing AI could reach the solution of one third tasks in ARC-AGI. This is due to excessive focus on LLMs, which are known to lack the ‘reasoning’ ability.

LLMs find it difficult to generalize since they are geared by memorization. They fail anything that is unrelated to their training data. What LLMs do is to memorize reasoning patterns, and cannot generate new reasoning.

Chollet and Mike Knoop arranged a competition to build open-source AI scoring high on ARC-AGI. About 17800 submissions were received. The best ones recorded 55.5 % score. They failed short of 85% human-level threshold. Many submissions where just brute force to a solution. The performance on puzzles was poor.

It is not correct to set ARC-AGI as benchmark since the very definition of AGI is not yet clear. AGI has already been reached if it is defined as AI ‘better than most humans at most tasks.’

In 2025, Knoop and Chollet will have another competition with another benchmark. However, defining intelligence for AI will be as controversial as it has been for human beings.

Bitcoin has risen in value after the Presential election in the US. It is expected that there would be crypto reserve supported by the government. There are crypto friendly legislators who too will find way to make this happen. It is difficult to see how the whole thing is beneficial for the US.

Bitcoin has some positive aspects. It is portable — being digital millions of coins can be carried on a pendrive. It has anonymity — holders are identified by an alphanumeric key. It can be easily transferred anywhere, with no intermediation of government banks and other financial institutions. It can be added to one’s investment portfolio for diversification.

Bitcoin cannot qualify as money — its volatility makes it unsuitable for this role since it is a poor medium of exchange. In most countries, it is not accepted for payment. Bitcoin transactions are slow and costly — they require compute power and energy for validation. Besides, you can lose your coins if you lose the thumb drive. Crypto is a different kind of asset. It is not money itself, but like money in some way.

Bitcoins do not attract interest or generate dividends. If the prices of the coins rise, there is no greater supply. The quantity of the coins is pegged at 21 million, and there are already 20 million Bitcoins. As such, if the demand rises, the prices could reach astronomical levels. The whole stock of Bitcoins at $ 99000 per coin is valued at $ 2 trillion.

Bitcoin lovers favour a government reserve fund in the light of this background. The government can purchase a certain quality of Bitcoins and hold it for a specified period of time. It is in addition to confiscated bitcoins from the criminals. This would escalate the price of Bitcoins, as investors would rush in before the government purchases the Bitcoins.

If Bitcoin is made a small part of an investment portfolio, say 2 per cent, the worth o Bitcoins would be in trillions.

How does it benefit the government or people not holding Bitcoins? Nothing. The government will be holding volatile tokens that do not generate any income. The treasury will have to borrow or print money for such an investment.

Cryptos could be supported by the government by formulating suitable laws and regulations. Cryptos could be made stablecoins supported by deposits at the Fed or treasury securities. There could be legislation to define whether tokens are currencies or securities. There should be rules for consumer protection. The use of cryptos must be banned for anti-social activities.

Crypto technology can facilitate the trade of financial assets or the transfer money across geographies. There should be guardrails to leverage the benefits of crypto technology.

ChatGPT has garnered 300 million users, and the company wants to quadruple this figure over the next year. Open AI may release its much-awaited video generation model Sora. It is likely to release its chain-of-thought (CoT) model o1. It can demonstrate its browser. In January 2025, OpenAI may release Operator, an AI agent.

Grog Brockman, one of OpenAI’s founders, was training DNA foundational models.

OpenAI has welcomed talent from its competitors, and seems to be on the lookout of veterans rather than raw talent in early twenties.

Google DeepMind released Genie 2, a large foundation model, capable of generating a variety of 3D environments. It can transform a single image into interactive virtual world. It facilitates development of embodied agents. It paves the way for new, creative workflows. Its predecessor was Genie 1. It was a 2D model. Genie 2 is 3D. It employs autoregressive latent diffusion technology. It generates sequential frames in response to user actions.

Google’s SIMA, an AI agent, performs tasks in Genie 2 following NLP instructions.

The model can generate new content on the fly.

Google is confident that it is marching better on a path to AGI than its competitors.

In 2025, digital ad spending is forecast to increase by 5.8 per cent (Dentsu Global Ad Spend Forecast report). The rise is attributed to AI-driven surge — AI placements of advertising. The report estimates that global advertising market will grow by 6.8 per cent year-over-year in 2024, closing at $ 772.4 billion.

America will lead the global ad spend in 2025, with a 6.3 per cent increase driven by digital and streaming investments. In EMEA, the growth is projected at 5 per cent supported by digital performance in the leading markets such as the UK.

Digital advertising will be the fastest growing channel, with a projected 9.2 per cent increase in 2025 and a CAGR of 8.8 per cent through 2027. By then, digital ad spend is expected to reach $ 513 billion, accounting for 62.7 per cent of the global market.

Algorithmic media capabilities provide an unprecedented opportunity to connect with consumers in more personalized and meaningful ways.

Social media advertising is projected to grow by 8.7 per cent in 2025, with a CAGR of 7.8 per cent through 2027.

Online video advertising is expected to grow by 8 per cent. Programmatic advertising will see an increase of 11.1 per cent in 2025, accounting for more than 70 per cent of digital ad spend with a CAGR of 10.9 per cent through 2027.

TV advertising will see a modest growth of 0.6 per cent in 2025. Connected TV will see the major growth due to ad supported streaming. Broadcast TV continues to decline (-2.5 per cent).

We have written several blogs on LLMs or large language models technology. Essentially, these models are trained to predict the next tokens or to guess missing tokens. However, they are not trained to accomplish tasks they are expected to perform. The training is expensive — billions of parameters and so many GPUs. The user ultimately pays the bill. The performance is derived from the heavy hardware around the models, and not from the neural networks themselves.

Therefore, there is research to improve the functioning of LLMs. Let us call it LLM 2.0. LLM 2.0 focuses on robust back-end architecture. It retrieves and leverages knowledge graph from the corpus (smart crawling). LLM 2.0 is hallucination free and does not require prompt engineering. LLM 2.0 has few tokens. It uses customizable PMI metric for key word associations. LLM 1.0, we know depended on vector databases, dot product and cosine similarity. (instead of PMI). LLM 1.0 relies heavily on faulty Python libraries for NLP. LLM 2.0 uses contextual tokens with non-adjacent words, sorted n-grams, variable key word associations, variable length embeddings, in-memory nested hashes.

LLM 2.0 focuses on conciseness and accuracy. LLM 1.0 gives lengthy English prose suited for novices.

In LLM 2.0, there are specialized sub-LLMs. LLM.1.0 does one prompt at a time. There is no real-time fine tuning. It is not based on explainable AI. LLM 2.0 uses multi-index and deep retrieval techniques. LLM 1.0 uses a single index. Proprietary and standard libraries may miss some elements in PDFs. It has shallow retrieval.

ChatGPT o1 works in a way that is closer to how a person thinks. It gives the hope that some form of AGI might be imminent, or even already there. LLMs rely on a method called next token prediction. The model is repeatedly fed samples of text broken into chunks called tokens. The last token in the sequence is hidden or masked. The model is asked to predict it. The training algorithm then compares the prediction with the masked token. It then adjusts models parameters to enable it to make a better prediction next time. The process continues, using billions of fragments of language, text or code till the model can reliably predict the masked tokens. At this stage, the model parameters have captured the statistical structure of the training data and the knowledge contained therein. The parameters are then fixed and the model uses them to predict new tokens. It is called inference.

The transformer architecture allows the model to learn that some tokens have a strong influence on others, even if they are widely separated in a sample text.

Chain-of-thought (COT) prompting improves LLMs by breaking dawn a problem into smaller steps to solve it. However, this process does not work for small LLMs.

Ambergris (literally means grey amber in French) is waxy substance generated in the digestive tract of the protected sperm whales. It is found floating in or on the shores of tropical waters and is used in perfumery as a fixative. The substance facilitates the passage of hard and sharp objects which the whales ingest while eating large quantities of marine species.

Mistakenly, it is called ‘the vomit of whales’. In fact, it is passed as faeces.

Freshly passed ambergris is light yellowish in colour and very fatty. As it ages, it becomes reddish brown. Sometimes, there are shades of grey and black.

It has a mild, earthy, sweet smell along with mild marine odour.

It is a coveted substance in international markets since it is very rare. Traditionally, ambergris was used to produce perfumes with notes of musk. These command high prices. Its use could be extended to flavour alcoholic beverages, food and tobacco, but it is rarely used for such flavoring these days.

Ambergris possession and trade have been banned in several countries (USA, Australia, India). Sperm whales are protected species under the Wildlife Protection Act in India.

Ambergris can fetch anywhere between Rs. 1 crore and Rs. 2 crore per kilogram in illegal markets.