Big Four accounting and audit firms in consultancy area have adopted AI whole-heartedly. EY has committed an investment of $1.4 billion to develop EY.ai which could be adopted by client orgnisations. Their AI LLM platform provides a secure environment to facilitate AI adoption by clients. PwC India has AI Lab in Gurugram. They have partnered with industry leaders such as Microsoft and OpenAI. Deloitte has unveiled a global AI market incubator in India to foster innovation. They have also tied up with Amazon Web Sevices (AWS) to drive generative AI-driven innovations for Indian businesses.

Consultancy firms such as McKinsey, BCG and Bain and tech majors such as Accenture and IBM are investing heavily in forming partnerships with IT companies and startups to develop solutions for their clients.

McKinsey has developed Lilli, an AI tool for consultants. They have partnered with Salesforce. The aim is to develop customised AI-driven consulting solutions.

BCG has developed BCG X in partnership with Anthropic to provide AI models such as Claude2 to clients. They have collaborated with OpenAI to offer customised solutions.

India has 867 commercial radio stations, including AIR, now called Aakashvani. It has phenomenal reach and is a hyper-local medium. Local retailers contribute 50-70 per cent to the radio operator’s ads.

India’s radio business has been severely affected by the pandemic, the emergence of streaming music and a host of structural issues.

Radio listenership and volumes of radio advertising have risen. However, its share in total advertising fell down, from 3.5 per cent to 2 per cent in 2020, and have stayed as it is.

Advertisers do not see radio as a strategic medium for frequency or reach. Radio operators have taken to event organisation, and this contributes 20-40 percent of their top line.

Telecom Regulatory Authority of India (TRAI) has made some recommendations to revive FM radio.

1.There are two fees for FM operator. One time non-refundable entry fee and licence fee. The licence fee is at 4 per cent of gross revenues (including GST) or 2.5 per cent of the one-time entry fee for a city. To illustrate, if the highest bid for a city is Rs.169 crore, the licence fee would be Rs.4.2 crore.

TRAI now recommends that licence fee should be delinked from non-refundable one-time entry fee. It should be calculated as 4 per cent of gross revenue (not including GST). If the license fees are rationalised, it will help the medium to become cost effective.

2.The government should support those operators who have been affected by pandemic.

3.Operators will be allowed to broadcast news and current affairs for 10 minutes per clock hour. This will bring advertisers.

4. Mobile phones should have their receivers active for FM reception. This should be made mandatory. Telecom companies disable the receivers on handsets to promote their own music streaming service. FM radio can reach over a billion mobile users. The current listenership can easily double.

Let us see whether the ministry accepts those proposals.

The G20 leaders were pragmatic enough and had foresight to see what a threat crypto assets posed to mainstream financial markets. The leaders were in favour of globally co-ordinated crypto regulations. They were also in favour of Reporting Framework for crypto assets, They want to have crypto exchanges for different jurisdictions as per co-ordinated timeline. They want oversight of crypto assets.

It is now clear that none of the G20 countries is inclined to give cryptos a legal tender status.

Cryptos, as we know, are decentralised digital assets. They have the potential of being misused. In fact, they have already been misused. However, this does not rob them of their convenience as safe haven for the investors. As we know, even regulated assets do not exhibit orderly behaviour.

In 2022, there was a crash of an FTX exchange, and Bitcoin suffered heavy losses. Later in 2023, there was resurgence. Regulation of crypto assets is not to legitimize them, but to minimise the risks and to respond quickly in the face of the risk.

G20 directions will accelerate the formulation of the regulatory framework. The RBI in India is in favour of banning the cryptos. It is at the instance of Supreme Court in 2020 that banking facilities have been provided to crypto exchanges. Policy makers want to provide an alternative digital currency (CBDC) here in India. That cannot rule out the co-existence of private cryptos along with the official digital currency.

Dollar has been the reserve currency of the world. Blockchain based currency might dent the dollar’s position as the only currency to settle global payments.

Emerging economies such as India are likely to benefit in cross-border settlements in their own currencies to counter dollar’s pre-eminence.

Gen AI can assist the programmers to write a code. All a programmer has to do is to articulate his problem. Generative AI also makes it easier to create product presentation before clients — the workflows, demo of products through videos and more. It brings quicker approval from clients. Gen AI can help write product notes and cases.

It is necessary to learn how to get the best from Gen AI by asking the right questions (prompts) to receive the best answers.

AI-powered coding assistants such as Microsoft’s GitHub Co-pilot and Google’s Studio Bot are a great help to the developers. It is easy to master them, and they easily integrate into development environments(IDEs) — Visual Studio Code, JetBrains and JupyterLab which are frequently used by developers.

The coding assistants are a great help in ML applications. They read a database table and apply algorithms such as SVM — support vector machines and gradient boosting trees to make inferences. The coding assistant finds the Python libraries and function calls. It can give a ready code to execute.

In not using coding assistant, you will have to import data, cleanse it, build training and testing datasets, evaluate results and create charts showing results. In the process, the problem being solved loses focus. A coding assistant writes 50-60 per cent of the code for the developer. These coding assistants support Python, C Sharp, JavaScript and other languages.

Real time coding suggestions are useful, and the developer has the freedom to accept or reject these.

Gen AI is a blessing to both the experienced and novice programmers. Experienced programmers break a programme for any problem into a set of sub-problems, the solutions of which are already available. It is a productivity booster.

Gen AI also converts the solution from one language to another.

Gen AI draws solutions from open source code, and may need fine tuning.

AI is both a threat and an opportunity for journalism.

London School of Economics surveyed over 100 news organisations from 46 countries about using AI and related technologies between April and July, 2023.

More than half of those surveyed (60 per cent) had concerns about the ethical implications of AI on journalistic values including accuracy, fairness and transparency. Journalism is undergoing metamorphosis on account of technology — the change is exciting, but at the same time scary.

Around 85 per cent respondents experimented with generative AI such as ChatGPT or Google Bard for different tasks — writing summaries and generating headlines. Almost 60 per cent had reservations. Journalists have recognised the time saving benefits of AI with tasks such as interview transcription.

Journalists acknowledge the need for checking AI-generated content by a human.

Creators in fact bring the audience on line. India is the fastest growing market on account of the creator economy in India. Creators to build multi-million dollar businesses. Creators in fact increase the user base of YouTube and generate more revenue for it.

Generation of users coming online is the key to growth. Some of these users have only their mobile phones to experience the online offerings. They connect with the communities and content on YouTube. Google engages with the creators both on a one-off basis, and on continuous basis. YouTube’s product teams have direct relationships with creators built over last so many years.

Those creators with millions of followers and high profile influencers must be on YouTube platform. This helps YouTube to compete with TikTok and Instagram.

YouTube wants to retain the existing pool of creators and wants to bring new creators in the eco-system. To this end, they are using generative AI. A new tool Dream Screen is being offered. It allows creators to generate videos and photos using AI to use in the background in their shorts. A creator wants to execute an idea — a dragon flying over Mumbai. They have to type a prompt, and AI generates the corresponding video.

There are AI-generated suggestions for the future videos. There is an automatic dubbing product to translate videos in other languages. There is an AI-assisted music recommendation system. It suggests music (audio) to use for the creator’s video.

At the root of Dream Screen are the foundation models. YouTube fine tunes some of these models so that creators can use them.

YouTube Shorts have become competitors to TikTok. These Shorts command 70 billion average views per day. However, it is an issue — how to monetise these Shorts. YouTube is investing heavily in Shorts. This is a part of their mobile first strategy.

In this data-driven world, predictive analytics put us on firm footing. It allows us to draw valuable insights and make rational decisions. Predictive analytics uses data, statistical algorithms and ML techniques to identify the future outcomes based on historical data. The aim is to see what the future would be like based on past data.

It is a process that uses data analysis, ML, AI and statistical models to identify patterns which predict future behaviour.

In short, predictive analytics is a technology to make predictions about what is unknown in the future.

There is wider interest observed in this field in the last five years. Another name for predictive analytics is advanced analytics. It has been linked to business intelligence.

If assesses risk, business trends and future maintenance requirements. In data science, they use various regression models and ML techniques to do this.

Predictive analytics lend a certainty to future, and this distinguishes itself from descriptive analytics.

It is useful in demand forecasting, production planning, insurance claims, software testing life cycle.

The following tools are well-entrenched in this digital age to do predictive analytics.

DataRobot

It is an automated ML platform used in predictive analysis. It enables data scientists to build predictive models. There are pre-built templates in it.

IBM Watson Studio

It is a comprehensive platform offering an array of tools for predictive analytics and data science. ML models construction and deployment could be done using its AutoAI feature. It has been integrated to IBM Cloud. It makes it scalable and flexible for all types of business — big, medium or small.

SAS Analytics

It has been around for quite some time. It keeps on evolving. It shows advanced analytics capabilities. It is useful in ML, model building and intelligence.

Tableau

We have already examined Tableau in detail in previous write-up. It is used for visualization. Its capacities have been expanded to cover predictive analytics. It has added features such as Explain Data and Ask Data. It makes available analytics to non-technical users. It has been integrated to cloud storage and ML libraries.

Google Cloud AI Platform

Google Cloud ML as well as cloud infrastructure. There is a suit of tools for Data Scientists. Model building becomes automatic. Being scalable, it is an ideal choice for organizations to leverage predictive analytics.

Some other predictive analytics tools are IBM SPSS (Statistical Package for Social Sciences), RapidMiner Studio, TIBCO Spotfire, H2O.

Some techniques of predictive analytics are decision trees, neural networks, text analytics, regression.

Just as movies require superstars such as Shah Rukh and Salman, the nation needs people who are good at mathematics. Those who contribute to nation’s progress are engineers and scientists. The jobs they perform involve a lot of mathematics. The challenges a country like the USA faces requires mathematics. Mathematics enables one to make calculations, analyse data and solve problems. Mathematics is becoming more and more a part of almost every career.

Of late, many Americans joke about how bad they are at mathematics. They show poor scores on standardised math tests. The advances in technology in the next 50 years are going to come from those countries who have intellectual capital. STEM education must be supported (STEM stands for Science, Technology, Engineering and Mathematics). China and Russia has many more STEM graduates. Such countries may challenge America’s dominance.

LLMs, as we know, are pretrained for natural language processing (NLP). Such pre-trained models have certain weights assigned to the tokens. The model is further trained to improve its performance for a new or for a specific task. The model can also be trained to adapt to a new domain.

LLMs are fine tuned on a new data set to make them improve for a specific task. The specific task could be translation or summarization or question answering. The pre-trained model has been trained on vast dataset, whereas for fine tuning we will use a smaller dataset relevant to the specific task, say question answering.

There are various techniques of fine tuning. The most common is the use of supervised learning. The model is trained on labelled dataset. In question answering, the dataset would consist of pairs of questions and answers. The model is trained to predict the correct answer of each question.

A model can be fine tuned by training it from scratch. However this is time-consuming and computationally expensive. A model can be fine tuned by freezing some of the layers of the model. Here we will not revise weights of the model. That prevents overfitting. In partial fine tuning, only a subset of parameters are adjusted.

Fine tuning is an effective way to make the model more efficient. However, fine tuning could also lead to over-fitting. Overfitting means the model has mastered the specific details too well, and hence it loses the capacity to generalize for the new data. To prevent over-fitting, we can use smaller learning rate, regularization and early stopping.

A large language model (LLM) is first of all an AI algorithm. It employs deep learning techniques and massive data in order to understand, summarize, generate and predict new content. A closely related term is generative AI. It is a type of AI that has been architected to generate text-based content.

As we know, language is a means of communication. Spoken languages are based on communications –both human and technological. These have evolved over several millennia. Language chiefly is a system of syntax, and uses words and grammar to convey ideas and concepts. A model of AI that is language model also serves the same purpose.

AI language models date back to Eliza which debuted in 1966 at MIT. All language models are trained on datasets, and deploy various techniques to infer relationships. Then they are in a position to generate new data. Of course, it is based on trained data. Language models are used for natural language processing (NLP). A user puts a query in natural language (input) to get a response in natural language (output).

An LLM is evolution of a language model — here the model expands the data based for its training and its capability to infer. There is no agreement on how large the dataset would be, a typical LLM has at least one billion or more parameters. Parameter here means the variables present in the model on which it was trained, and which it can use to infer new content.

The genesis of modern LLMs goes back to 2017 when transformer models came up as neural models. LLMs thus have a large number of parameters and transformer architecture. LLMs have applications across many different domains.

LLMs are referred to as foundation models. This is a term coined by Stanford Institute for Human Centred AI in 2021. A foundation model is large and impactful. It is a foundation for further optimizations and specific use cases. AI and ML are the techniques to enhance efficiency (input/output ratio), effectiveness ( more output per unit of input), experience and evolution of an organisation. It is because of these benefits that businesses are inclined to invest in this technology.

Working of the Language Models

As we know by now, an LLM model is trained on a large volume of data called corpus. The dataset is generally expressed in terms of petabytes. There are steps in training. First, there is unsupervised learning where the model learns on unstructured data and unlabelled data. Its advantage is the vast data that is available . Here the model gets the capability to derive relationships between words and concepts.

As a next step, LLMs are fine tuned on self-supervised learning. There is some data labelling. It assists the model to identity the concepts more accurately.

Further, LLMs undertake deep learning through transformer neural network process. It enables the model to grasp the relationships and connections between words and concepts using self-attention mechanism. A score is assigned (commonly called a weight) to a given item or token in order to decide its relationship.

After the training process is over, there is a base for AI. If a prompt is given as a query, the model generates a response, say an answer to a question or some new text or summarised text or sentiment analysis.

Uses of LLMs

LLMs are used for text generation, translation, content summary, rewriting content, classification and categorization, summary sentiment analysis and conversational AI and chatbots.

Challenges in Building LLMs

There are huge development costs since we require expensive graphic processing units and massive data. There are operational costs which are high. A bias could sneak into the data. The model could be a blackbox, and we are unable to explain how it arrived at particular decision. There are inaccuracies or hallucinations. LLMs are complex models with billions of parameters. If a prompt is malicious, there are glitch tokens causing malfunction of the LLM.

Types of LLMs

There are generalised zero-shot models, and there are fine-tuned domain specific models. There is a BERT model or bidirectional encoder representations for transformer model. Lastly, there is multi-modal model that handles both text and videos, say GPT4.

Future

Models can acquire artificial general intelligence or become sentient. Models use techniques such as reinforcement learning from human feedback (RLHF). Google uses Realm or retrieval augmented generation language models.