Blog

Elon Musk presented a lithium-ion battery powered aircraft that flies in a single charge up to 1000 kms in the last week of February, 2025 at California’s private airfield. The aircraft is called Tesla Air. Its body is made up of ultra-light carbon fiber to allow it maximum aerodynamic efficiency. Its wings are designed to increase the lift. Its propulsion system is fully electric. Its battery can be fast recharged in less than 45 minutes.

Aviation industry contributes heavily to carbon emissions. Tesla Air heralds clean and quiet air flights.

The cabin interiors are minimalist but comfortable. The aircraft is a turning point for the future of air transport.

There are some challenges. However, by 2040 most regional flights could be operated by electric aircraft.

The sky is not the limit for Tesla.

The present time happens to be the most opportune time to avail of emerging technologies, and at the same time the most uncertain time. There are major shifts in the external environment — market dynamics, workforce transitions, economic slowdowns, AI agents and geopolitical tech rivalries.

At the same time, there is a high demand for talent.

This is a period of rapid change. Smaller organizations attract talents which once gravitated to the industry giants. As technology permeates every aspect of our life, there is a need for tech expertise. Despite uncertainty, technology is the driving force behind innovation, growth and competitive advantage.

At this inflection point, there is convergence of opportunity, uncertainty and necessity.

Modern technologists cannot be put into silos — software developer, AI researcher or data scientist. These job descriptions fall short of the evolving expectations. Industry requires professionals who can blend technical skills with design thinking, ethics, human-centered innovation and policy awareness.

Skill development is concerned with the change seen today. It is not for future preparedness. There should be investment in high tech skills — AI, analytics, automation and beyond. Mere technical expertise will not take you further. There has to be an ability to adapt, think critically and solve inter-disciplinary problems faced by the next generation.

There are individual success stories of Indian tech revolution. However, India needs an ecosystem unifying these innovations.

India has introduced UPI — a digital payment system. It is an open, interoperable, inclusive model. Similarly, instead of replicating western models, India should carve its own path leveraging its strengths to build an ecosystem that is robust and unique.

India has shown its readiness to embrace technological advances — AI Mission, Semiconductor Mission and Quantum Mission. There should be long term strategy for sustained impact.

There is a race for global tech dominance. Sucess will not be measured by individual achievements. Success will come to those who integrate technology and technologists into governance, industry and society.

India’s music labels are suing OpenAI for ‘unauthorized use of sound recordings’ in training AI models. In doing this, OpenAI has breached copyright. Music companies have joined ANI which has accused OpenAI of using its news content to train ChatGPT without permission. The Federation of Indian Publishers has filed a similar suit against OpenAI at Delhi HC. OpenAI also faces a case from media outlets for improper use of copyright material.

Section 14 of the Copyright Act, 1957 provides owners of exclusive rights, including rights to reproduce the work and communicate it to the public. The courts will have to decide if rights enumerated in Sec 14 are violated by using copyrighted material to train AI models. There are exceptions to infringement provided under the Act. Whether they could apply in this case is to be decided. One such exception is that of fair use that permits limited use of copyrighted material for private/personal use, criticism or review and reporting of events.

OpenAI claims that Indian courts lack jurisdiction since the company is US-based. It also claims that it follows fair use principles in deploying publicly available data to create its AI models.

It is now for the courts to decide how to regulate AI without stifling innovation. The case could set legal precedents — whether AI-generated content qualifies as copyright infringement and whether developers must seek prior permission to use copyright material.

The UK government last year proposed to let AI developers use copyrighted content without permission and strike licensing agreements between AI firms and content creators.

US AI startups have invested tens of billions of dollars into AI models in the hope of a big pay off in future . DeepSeek emerged out of a quantitative hedge fund run by founder Liang Wenfeng and has no pressure to build a revenue model — just pure garage-energy and community-driven innovation.

There is a tiny team exploring AGI. In the last week of February 2025, they will be open-sourcing 5 repos. A code and data repository is digital storage space where data and resources needed to operationalize AI models are organized and managed — training, running and evaluating AI models. The Hangzhou-based startup’s technology has been fully tested, deployed and documented.

It intends to go further by publicizing the underlying code, the data used to create it, the way it develops and manages that code.

It is seeking wider adoption of its technology.

Since Jan 23, 2025, DeepSeek, a low-cost foundation model from China, shook up the AI industry. It is an open-source model that has been built at a fraction of the cost of its US competitors. That sent Nvidia stock on a downward spiral. DeepSeek was trained on inferior GPUs compared to ChatGPT. The event also triggered an export ban on the Chinese models. In India, there is a discussion whether the country is losing in the AI race for want of a locally developed LLM.

As part of its AI Mission, the Centre has received 67 specific proposals to build an India-specific AI foundation model, including 20 LLMs. A high-level technical committee will evaluate these proposals.

The proposals for LLMs are received from Sarvam AI, CoRover.ai and Ola. There are 20 proposals for LLMs. The remaining are sector specific LLM proposals.

We have seen that traditional computing uses binary computing or bits, which process data in 1s and 0s. Thus, any calculation can be done only in one set of numbers at a time. Quantum computers, on the other hand, use qubits which exist in multiple states at once, making them exponentially more powerful for certain tasks. There is no limitation of sequential computing, and several tasks can be done simultaneously.

Google and IBM’s quantum processors focus on increasing the number of qubits. There is a major limitation here of correcting errors. What we need is error-corrected qubits. Google achieves this through increasing qubit counts and reducing errors through improved hardware and error-correction techniques. Microsoft has taken a different approach. It has introduced Majorana 1 quantum chip. It is a new type of qubit that uses a special kind of particle that makes qubits more reliable. It means fewer errors and greater potential to scale quantum computers to the size needed for real world applications. Microsoft aims to solve the problem at its roots by developing qubits which are more stable.

Quantum computing is not just a futuristic idea. It can revolutionize industries by tackling problems which traditional computers cannot tackle. The combination of quantum computers and AI will make AI models more efficient. Of course, one area of concern is that quantum computing can easily break encryption technology currently in use to protect data and this poses a serious security risk.

Quantum computing is not a replacement of traditional computing. It can supplement the existing technologies, and could be reserved for specialised applications.

Microsoft experts intend to build a useful quantum computer between 2027 and 2029. The competition is heating up. Each advance brings us closer to a major transformation.

The other challenge is the integrity of the outcome. The multiple states of a qubit lead to multiple outcomes, only one of which is desirable. getting one desirable outcome instead of myriads of possibilities is also a challenge. There are errors in calculations due to disturbances. Quantum behavior of a particle collapses into normal behaviour, the moment the system is observed or measured. Besides the are disturbances such as change in temperature or presssure which makes the system collapse. Maintaining the stability of qubits is a huge issue.

xAI is the AI startup set up by Elon Musk. It has released Grok-3, the latest iteration of its chatbox. It will directly compete with DeepSeek, ChatGPT and Gemini. Musk moves aggressively to expand xAI’s influence.

The chatbot is available to premium subscribers on X. xAI is also launched a new subscription tier — SuperGrok for mobile users and website users.

Grok-3 is in a league of its own. It outperforms its predecessor Grok-2.

Grok-3 puts xAI in the race for leadership in open source LLMs.

xAI is strengthening its data center capacity to train more advanced models. It has a supercomputer cluster in Memphis called Colossus. The latest release comes with a smart search engine Deep Search, a reasoning-based chatbot.

Grok-3 introduces features such as Think Mode which provides step-by-step reasoning for users. There is Big Brain Mode which uses more computational resources to handle complex tasks.

The name Grok is inspired by the character Grok in Robert Heinlein’s science fiction novel Stranger in a Strange Land. The character was raised in Mars and is meant to convey deep understanding and empathy.

Grok-3 can be accessed on X (formerly Twitter) by subscribing to X-Premium+ service or SuperGrok tier. It costs $30 per month. In India, it is available for Rs. 3470 per month. Grok-2 can be accessed free of cost, and the company is planning to make it open source.

Grok-3 was trained on 2 lac Nvidia H100 GPUs in two stages — initially with 1 lac GPUs followed by further training of 92 days scaling up to 2 lac GPUs. It could have cost $6 billion-$8 billion. Grok-2 was trained on 1 lac GPUs.

Audience is having a wait-and-see approach. There are misgivings about its practical usability. Its responses may be too lengthy or lack in clarity.

Artificial intelligence has witnessed several waves of technology. The chief scientist of AI, Facebook, LeCun was fascinated by intelligence in animals and humans right from his young age. Later, he worked on neural nets. While accepting Facebook assignment, he insisted on working on open-source models. He also insisted that he would be New York-based and would teach at NYU. He considered that open source makes AI democratic enough for the next wave. He remembers Minsky who talked about the perceptron in the 1950s. He refers to the work of Piaget and Chomsky in linguistics. He is aware of the limitations of data and learning models of the 1980s. Minsky and others felt that it was time to pause AI research (AI winter). The slump in AI research lasted for many years.

LeCun is not fond of the term Artificial General Intelligence (AGI). He prefers the term AMI or advanced machine intelligence. There is no linear scale of intelligence, say a cat smarter than a human being or vice versa. Some gadgets can beat an intelligent human being in some games, say chess. AGI is not going to be an event. It is a progressive thing.

AI will be smarter in human ways in long term. However, there is no single tipping point.

LeCun is of the opinion that generative AI could be replaced by Joint-Embedding Predictive Architecture (JEPA). GenAI has a short shelf-life.

He mentions PyTorch as the vehicle for ChatGPT and other technologies. He feels that self-supervised learning is probably the most revolutionary concept that completely changed the way we practise ML. In future, our interactions with the digital world will be mediated by AI assistants.

Printing press was the last invention that accelerated dissemination of knowledge and philosophy. It led to American revolution, the French revolution and emergence of democracy. AI could have such a profound impact, and will bring about a new renaissance.

DeepSeek, the Chinese chatbot, made its debut on Monday, the 27th of January, 2025 sending a shockwave throughout the Silicon Valley. Nvidia lost almost half a trillion dollars in value on that fateful Monday. Nvidia has since then recovered, and its revival has been achieved by the appearance xAI’s Grok-3. Grok-3 has a huge stockpile of 1 lac GPUs, which could expand to a million chips. xAI is progressing at superhuman pace. This proves that the route to success in cutting-edge AI runs through Nvidia and this will not change in near future. DeepSeek stimulated US efforts to build more efficient smaller models, but it is not a threat to the spending on Nvidia chips.

There are competitive pressures, e.g. Amazon.com chips made inhouse such as Trainium and Inferentia, which are more economical for training and running AI models. However, Nvidia is not just selling chips but an AI platform – CUDA which is popular among AI developers and is compatible only with Nvidia chips.

Investors may overreact in future too about any depressing signal. There are snags in Nvidia’s Blackwell chips. There are export curbs on Nvidia as far as China is concerned. However, these may not throw Nvidia massively off course. There are issues of AI being overhyped but that will be known in future. Till then Nvidia chips will be bought.

At the AI Summit in Paris, the US and UK refused to sign a declaration advocating for an ‘open, inclusive and ethical’ approach to AI.’ The declaration was endorsed by 60 countries including India, France, Germany and China.

Since the release of ChatGPT in 2022, there is a huge flow of capital into equites, this leading to an increase of around $10 trillion in the market value of the ‘Magnificent Seven.’

The US has 61 LLMs, followed by China (15). The release of DeepSeek has intensified the rivalry between China and the US.

There is a global power struggle over AI. Europe wants strict rules and public funding. China is expanding state-backed AI. The US has a free-market approach.

Generative AI has revolutionized the AI landscape. There are discussions on national security and data protection. There is no agreement on AI regulations. The is no agreement on AI regulations. It is necessary to have a global vision for AI. A global framework is desired so as to enjoy the advantages of AI and mitigate its risks.

The US and Europe have opposite views on regulating AI. The US thinks its AI technology is the gold standard worldwide. Its advocates least government intervention. Europe feels AI should be safe.

,