Unpacking What It Takes To Build Something Like ChatGPT

There are few things in today's fragmented world that can reach everyone, whether it is news, music or art. ChatGPT has been able to reach every corner of the world. What does it take to create something so impressive? Why is it hailed as unmatched?

Srikanth Vellamakanni, cofounder and group CEO of Fractal, at the Entrepreneur India's Tech & Innovation Summit in Bangalore recently concluded, cited the introduction of Siri, by Apple, as an example to address the elephant in the living room. Apple's voice assistant, Siri, was first integrated into the iPhone 4S in October 2011. It took a few years.

"Siri made some hilarious mistakes. People thought that it wouldn't work. Sundar Pichai, just five years later, announced that 20% of all Google searches are voice searches. The technology has changed a lot over the last five years."

He explained how this happened by saying, "In 2020 Sam Altman CEO of OpenAI made a statement stating that data was important but that what's even more important is how much computation you are throwing. In the world of Artificial intelligence (AI), one of the key equations is that the error rate is dependent on data, computation, and the techniques you use. The higher the data quality, the more valuable it is. "If you combine a lot of data with a huge amount of computation, along with the correct techniques, you can reduce error rates."

Siri and ChatGPT both improved accuracy. The accuracy improved due to the multiplicative effects of data, computation and techniques. In 2011, Steve Jobs introduced the iPhone 4S and Siri. The error rate in voice recognition was 70%. Siri was therefore unable to understand. Now, the error rate is less than 4% which is higher than human accuracy. Adoption really took off. ChatGPT has reached 100 million users within a very short period of time.

JavaScript and Python have been the most popular programming languages in the past three years. Velamakanni claims that today, English is the most popular programming language. Velamakanni claims that code is generated automatically if English is written. This is GitHub Copilot. This is a dramatically faster tech. "You suddenly realize you can be 200x more productive," said Velamakanni who co-founded Fractal 2000.

The company has raised over $300 million in institutional private equity investments. Fractal provides AI, data, and analytics-driven strategies to more than 100 Fortune 500 Companies. The company has created several product lines including Crux Intelligence and Eugenie.ai.

The IT industry produces 10 lines of code a day. Imagine what the IT industry will do in seven hours if they write 10 lines of code within the first minute.

The obvious question is whether AI will eventually replace the job.

Velamakanni is of the opinion that AI will not replace humans. What will happen is that people who use AI are going to replace or outperform those that do not. The key point to remember is that it tells us more than ever that humans who use AI will be better off than humans who don't. This AI represents a paradigm change in computing. In the past 100 years of computing, machines were programmed to faithfully reproduce what was told to them. He said, "That has changed."

Today, machines can be taught to think. In supervised learning you provide examples, and in unsupervised you simply give a lot of data, and it will figure out anomalies, patterns etc. In reinforcement learning, the robot performs actions and receives rewards. It gets better at it. ChatGPT's self-supervised learning is the latest thing, and it uses old data to learn.

He said that it takes a sentence and attempts to predict the word after the last one. By predicting and learning the next word, it receives feedback and learns. It's a type of self-supervised, or unsupervised, learning that is creating the magic in AI.

AI can match and exceed human performance in many areas, including image recognition, speech detection, games, breast-cancer diagnosis, etc. In each of these fields, we're very close to the point where a machine generalized can begin to match or exceed human performance. He said that it's not a single task but rather a wide range of them.

Fractal’s mission is for AI to be used to assist Fortune 100 companies and power human decisions in every enterprise.