Blog

Scientific research is documented through research papers published in peer-reviewed journals. A large number of papers — somewhere between 2 million and 6 million — are published every year. It is estimated that at least 2 per cent of papers are fake — this number adds up to a lot.

These fake papers are churned out by the so-called paper mills. These papers have either full data fudged, or part of the data fudged. The paper mills operating in the field approach the scientists offering to write papers with credit given to the scientists for a price.

Paper mills have proliferated as the quantity of research is rewarded, rather than the quality of research. These paper mill studies get cited in legitimate review papers if the review writing authors are not careful and are after volume of work.

The funding agencies are impressed by the bolstered resumes of such scientists and precious resources are routed to such scientists, rather than genuine scientists.

Some papers are generated by AI. These days they use tools like ChatGPT.

There is no rigorous evaluation system for these papers. The evaluation is done by recruitment committees or grant committees. They do not have the wherewithal to make an actual evaluation. Scientists are rewarded on the basis of the number of papers they write and the number of publications that cite them.

Even legitimate research papers do not advance the state of knowledge. Researchers add some additional data in an ongoing project. It is converted into a new paper. Majority of these papers make no contribution. They are not worth reading.

The fake papers use a template. Only data and words are filled. Paper mills fabricate papers in those fields which tend to be formulaic (nano technology, computer science and mRNAs).

Many a time, the fake papers are retracted. However, their impact persists as these are cited and mentioned in review papers.

Funding should be made available rigorous peer-review.

Apple’s Vision Pro headset shows extended reality. It comes in handy in health space. To get the improved outcomes, surgeons are using the device from bariatric surgery to spinal surgeries. In a bariatric surgery, the device offers an immersive 3-D environment allowing the medical team to have a view of the complicated anatomical structures with clarity. It is a significant milestone in the journey to revolutionize surgical procedures. Apple Vision Pro have been used for laparoscopic surgeries. Simultaneously, a surgeon can see a CT scan in the device itself. The view of patient’s internal organs can be enlarged to the size of a wall, and they could be seen in greatest detail.

The device can be used to facetime with experts to get their opinion while carrying out the procedure. The device can be used to teach medicine.

The device has been used for a shoulder arthroscopy and a delicate spinal surgery. In the spinal surgery, the nurse wore the device, rather than the surgeon, offering critical support. The device has also been used in shoulder joint replacement procedure.

Apple headset is designed such that one can toggle between the real and digital worlds. It has been priced at $3500. It spans across communication, entertainment and more.

Media has, of late, covered weight loss drugs extensively. It has created a buzz around them. There is such a rush for weight loss drugs that this market is estimated to reach $130 billion by 2030. (Goldman Sach’s Research).

Among the latest formulation is Novo Nordisk’s Ozempic and Wegovy. Basically, they are anti-diabetic drugs used for weight loss. They increase insulin secretion, and weight loss is an added benefit. Chemically, these are glucagon-like peptide 1 agonists, which mimic hormone GLP 1. (The hormone is produced in the gut and signals the brain when a person is full). Ozempic and Wegovy are semaglutide anti-diubetic medication.

Semaglutide 2.4 mg makes you lose weight to the extent of 15-25 per cent.

Eli Lilly’s has introduced Moujaro and Zepbound . Maujaro’s active ingredient is trizeptide. It is GLP-1 and GIP receptor agonist. There are injectables of the strength 5 mg, 10 mg and 15 mg. It causes a weight loss of 15-20 per cent.

It is an impressive quantum of weight loss. These medications are not prescribed to those with a history of pancreatitis.

Another more effective drug that is undergoing trials is Retatrutide which acts via three different receptors and pathways and causes a weight loss of 25 per cent in 48 weeks. Its weight loss is equivalent to what can be achieved through bariatric surgery.

These medicines are not available in India and would not be any time soon.

There is global supply shortage too because of heavy demand.

Buyers are turning to flourishing gray market to get the supply. They try to get imported bulk packs and alternative medications.

Doctors report here in India women largely between their 20s and early 50s come seeking prescriptions.

Novo Nordisk markets Ozempic and Wegovy and earns tens of billions of dollars. Novo is a Danish drug maker. Eli Lilly started selling Zepbound since December 2023.

Obesity drugs are a global phenomena and often in short supply. Novo has no plans to launch Ozempic in India but is working to make Wegovy available.

What is available in India is Novo Nordisk’s Rybelsus pill, which contains the same ingredients (as Wegovy and Ozempic) but comes in the tablet form.

Apart from drugs, we have to focus on diet and exercise to get rid of obesity. Drugs when stopped make you gain weight again. In addition, after weight loss for six months, there is a plateau.

As per a Lancet study, India has third largest number of people with obesity (2022) behind China and the US.

After the advent of LLMs and ChatGPT, there are huge opportunities in generative AI space. Big corporates and startups are trying to take advantage of these opportunities. The generative AI market between 2021 and 2023 and the generative AI startups in India have doubled. The market will show an annual growth rate (CAGR) of 27.66 per cent between 2023 and 2030.

NeuralGarage is an AI startup. It operates in audio-visual (AV) space. It changes the lip and jaw movements visually using audio as a controller. It is a studio quality change. A TV commercial shot in one language (say Hindi) can be recreated into another video which looks as if it has been shot into that language (say Tamil). Thus, one can create multiple videos by shooting just one video with an actor/model. Their client list includes brands such as Amazon, Britannia, Coca Cola, Dream 11, HP, Ultra- Tech Cements, Eno and Microsoft.

This startup has also started a voice cloning service. The voice of a dubbing artist can be changed to the voice of a celebrity, another actor or model.

Neuropixel-AI is a startup that automates the apparel cataloguing process. It can generate synthetic models and improve virtual try-on experiences.

A chatbot like ChatGPT is Gupshup. It has launched a bot building tool — Autobot Builder based on GPT-3. The company built Tia, for Tata Capital to bring contextual and multi-lingual assistance to customers.

Startup companies help the legacy companies by offering generative AI tools that seamlessly integrate with their existing tech stack. Co-rover offers virtual assistants that streamline interactions. At the same time, they do product recommendations, help transaction processing and address customer concerns.

Generative AI facilitates the recruitment and selection function of HR. There are tools such as AI Screening and AI Interviews. These make most suitable candidates move forward by evaluating multiple attributes.

Gen AI solutions are offered to financial institutions. There are personalized recommendations and conversational commerce. These make available the suitable financial services to the customers by taking into consideration the context and the needs.

The pricing model followed are pay-as-you-go or subscription. In future, there could be revenue generation from customization and training.

India is taking efforts to build its own AI system by investing in computer infrastructure from scratch, along with cultivating AI talent and developing and deploying LLMs. Compute infrastructure may not come cheap. As such, the government has recently approved a fund of Rs.10371.92 crore which could be further raised to Rs. 20,000 crores.

India joins Saudi Arabia and UAE, and many other countries in the global race to shape the future of AI technology including Britain, France, and Germany. Interest in AI has reached sky-high since the advent of ChatGPT introduced by Microsoft-backed OpenAI in November 2022.

Silicon Valley has taken the lead. But AI nationalism is an all-time high elsewhere. Startups are now developing nation-specific models.

Dell computer has set up an AI factory to accelerate enterprise and AI factory to accelerate enterprise and AI integration in collaboration with Nvidia.

Ilaya Sutskever, a scientist and Jan Leike, an ML researcher from Open AI resigned in May 2024. They were in a team whose job it was to make sure humans are safe from OpenAI’s superintelligence. It is not known whether they will be replaced by other people.

Ilaya Sutskever, in fact, is a cofounder of OpenAI. Of course, while quitting he admitted that it was an honour and privilege to have worked together with Altman and crew. He expressed the confidence that OpenAI will build AGI that is both safe and beneficial.

Jan Leike’s departure was more abrupt.

Both were members of the super alignment team at OpenAI. Though mankind requires scientific breakthroughs to steer and control AI systems smarter than human beings, there is a need for a team of super alignment (July 2023). The team will ensure that AI systems much smarter than humans will follow human intent. OpenAI has recognized in July 2023 that there are no controls in place. The intended superintelligent AI could go rouge. The current RL HF relies on humans’ ability to supervise AI. However, it is a moot point whether humans could supervise AI systems much smarter than them. The current alignment techniques may not scale to superintelligence. There is a need for new scientific and technical breakthroughs.

Prior to his employment at OpenAI, Leike was working with Google’s DeepMind. He was dedicated to keeping humans safe from the superintelligence. Leike stated the alignment problem like this — when machines do not act in accordance with the human intentions. This problem has to be solved, and we should see what is needed to solve it. Leike wrote in March 2022 whether the alignment problem is located in the space humans can solve or cannot. The effort to solve the whole problem can lead us to something we cannot reach. A less ambitious goal can lead us ultimately to a solution — a minimal viable product (MVP) for alignment.

Agencies today have two businesses — the traditional media agency and the digital agency. Many agencies have started investing in digital marketing since the beginning of the new millennium, even though the client spending on digital is not significant then.

The employees recruited previously had read all marketing, advertising and brand management textbooks. It is true that there was no digital marketing textbook then. In a sense, the new age agencies are writing the textbooks for the future by creating knowledge and expertise that can be replicated.

A major chunk of business still comes from the traditional media business. Though new age marketing disciplines add a smaller volume to the business, that is the more exciting growth story. The new business did not exist a few years ago. Still, it creates differentiation.

The traditional business is being automated and is being made process-driven.

The agencies today recruit data scientists and content creators. They make investments across e-commerce, data acquisition, enhancement and enrichment and content. A large chunk of digital work originates from clients in e-commerce, data and content. In terms of volume, it may not be huge. But it is still exciting. Brands are also willing to pay more to ensure effective advertising.

Reverse-mode automatic differentiation is the mechanism underlying the backpropagation algorithm used in training neural networks.

Automatic differentiation (AD) numerically evaluates the derivative function specified by a computer programme. It leverages the chain rule of calculus to compute derivatives systematically.

There are two modes of AD. Forward mode computes derivatives alongside the function evaluation. It is suitable for function with fewer inputs than outputs. Reverse mode AD computes derivatives by performing first the forward pass to evaluate the function. Later, a backward pass is performed to propagate derivatives from outputs to inputs. It suits for functions with fewer outputs them inputs.

It is particularly more effective in neural networks as they have a large number of parameters (inputs) and fewer loss values (outputs).

Backpropagation is an application of reverse mode AD. It works by performing a forward pass to compute the output. Loss is calculated. Then a backward pass is performed to compute the gradients of the loss with respect to each parameter in the network, using the chain rule.

Backpropagation, in a nutshell, is a specialized use of reverse mode AD customized for computing gradients in neural networks.

TCS has launched a new platform. TCS AI WisdomNext in June 2024 to accelerate the adoption of generative AI technology. As we are aware AI has been around for quite some time, and generative AI is the new kid on the block. Organizations were getting acquainted with this technology in 2023 — where to use it and how to use it and how beneficial it is.

Scaling of this technology will take time just as cloud technology took time to reach a scale. The adoption of generative AI will be faster, but still, it will take some time.

Generative AI companies are moving up the value chain — from proof of concepts to production.

The new TCS platform offers pre-configured blueprints (with in-built guardrails) for industry specific solutions. These can be used to design and deploy enterprise AI solutions much faster.

The customers can unlock the full potential of their data, drive greater business innovation and efficiency and gain a competitive advantage.

The technology can be harnessed to solve business problems. The platform was showcased to over 100 customers. Different business sectors such as consumer goods companies, travel, BFSI, retail, manufacturing, life sciences, healthcare have evinced interest.

There are two challenges in the implementation of generative AI — technical (100 ways of implementing or 1000 use cases) and change management (where should the data reside, who is able to do what with the data and what kinds of models will get trained). Both these challenges will be addressed in due course.

TCS has already trained more than 3 lac employees in AI.

It is estimated that generative AI’s current contribution to IT companies revenue is between 1percent and 3 percent of the total revenue. This is likely to go up in the coming years.

Conforge has developed an AI platform called Quasar which is powered by 23 LLMs such as GPT series (OpenAI), Gemini (Google), and LLaMA (open source). Quasar offers six accelerators — Quasar Document AI, Quasar Speech AI, Quasar Vision AI, Quasar Graph AI and Quasar Conversational AI. All these are available on Microsoft Azure marketplace.

Quasar has 100 solutions and capabilities built into it. It has a collection of over 100 APIs. It has a library of 100 plus pre-built cognitive and generative use cases.

It offers capabilities across various domains. It allows leveraging the models of users’ choice — it includes document processing, speech recognition, development of predictive and prescriptive models.

They see great adoption. It offers contact center automation for banks.

It helps onboarding both customers and employees. It is a great help in training employees.

Is it necessary for the IT firms to build their own LLM? According to Conforge, the answer is a ‘no’. It is a huge investment. And IT company deals with customers from different industries — diverse domains. It is not practical to build an LLM from scratch for a specific domain. They, instead, use open source LLMs such as LLaMA. They invest in fine-tuning and retraining these models for specific purposes.