Unpacking What It Takes To Build Something Like ChatGPT Srikanth Velamakanni, co-founder and group CEO, Fractal talks about ChatGPT and the future of AI at the recently concluded Entrepreneur India's Tech & Innovation Summit 2023 in Bangalore
By S Shanthi •
Opinions expressed by Entrepreneur contributors are their own.
You're reading Entrepreneur India, an international franchise of Entrepreneur Media.
In today's world of fragmented consumption, be it news, art or music, there are very few things that reach every person across the world. ChatGPT is one such thing that has reached the nooks and corners of the globe. But, what it takes to build something like that? And, why is it being lauded as unparalleled?
Addressing the elephant in the room, Srikanth Velamakanni, co-founder and group CEO, Fractal at the recently concluded Entrepreneur India's Tech & Innovation Summit 2023 in Bangalore, quoted the example of the introduction of Siri by Apple. The Apple voice assistant was integrated into iPhone 4S back in October 2011. However, it didn't take for few years.
"Siri was making some incredibly comical mistakes. And people thought it would not work. But just five short years after that, Sundar Pichai announced that 20 per cent of all Google searches were voice searches. Technology has evolved a lot in just those five years."
Explaining how that happened, he said, "In 2020, Sam Altman, CEO of OpenAI, made a statement that data is important, but what is even more important than data is how much compute you are throwing. So, one of the most important equations to understand in the world of Artificial Intelligence (AI) is that error is a function of data, compute, and the techniques that you are bringing to the part. The more, the better, the higher the quality of data, the better it is. If you bring in lots of data, and then throw in an incredible amount of compute, and with the right techniques, together, you can bring the error rates down."
The same happened with ChatGPT and Siri. The multiplicative effect of data, compute and techniques improved the accuracy. "Back in 2011, when Steve Jobs launched iPhone 4S with Siri, the error rate was 70 per cent in voice recognition, therefore, it was not able to understand. Now the error rates are in this sub 4 per cent which is better than human accuracy. Thus the adoption really took off. That is one of the reasons why ChatGPT reached 100 million users in such a short time," he said.
Hottest programming languages
In the last three years, the hottest programming language network has been JavaScript and Python. Today, it is English, claims Velamakanni. "Because now if you write English, code gets generated automatically. This is GitHub Copilot. This is dramatically accelerating tech. You suddenly realize that you can be 100x 200x in terms of productivity," said Velamakanni, who co-founded Fractal in 2000.
The company has raised more than $300 million from institutional private equity investors. Fractal serves more than 100 Fortune 500 companies with its AI, data and analytics-driven business strategies. The company has developed several product lines within the company including Crux Intelligence, Eugenie.ai, Theremin.ai, Qure.ai, Samya.ai and Senseforth.ai.
"The productivity of the IT industry is 10 lines of code per day. Take it by 10 lines of code in the first one minute and imagine what they are going to do for the next seven hours," he added.
Thus, the obvious and most question today is if AI will replace jobs.
Will AI replace human beings?
Velamakanni believes that AI is not going to replace human beings. But what's going to happen is that humans who use AI will replace or outperform humans that don't use AI. "The key thing to take out is that this tells you more than ever, ever before that humans that use AI will outperform humans that don't use AI. This AI is a paradigm shift in computing. Because if you think about computing for the last 100 years, it was about telling machines what to do and machines faithfully reproducing what you tell them to do. That has changed," he said.
Machines can today be trained to learn to think. "In supervised learning, you give examples and in unsupervised learning, you just give lots of data and it figures out anomalies, patterns, etc. And in reinforcement learning, it does actions, it get rewards, and it gets better and better. As the new thing that has come up, which ChatGPT has done, is self-supervised learning where it learns with the old data," he said.
It, he said, takes a sentence, and tries to predict the next word in the sentence. And by predicting the next word and learning it gets feedback and learns from it. "It's a form of self-supervised learning, which is creating all this magic for AI.
AI is matching and exceeding human Performance in a broad range of tasks such as image recognition, speech recognition, games, breast cancer diagnosis, others. "In each of these areas, we are very close to that place where a generalized machine can start to match and exceed human performance. It's not just a narrow task, but a broad range of tasks," he said.
Fractal's mission is to power every human decision in the enterprise and to use the power of AI to help Fortune 100 companies.
