Why GenAI is the key to greater innovation

Until a few years ago, Generative AI or GenAI was a buzzword: today, it's the norm. In an era where staying ahead of the curve is critical, this technology is emerging as a powerful catalyst.

Businesses are changing the way they work with GenAI, increasingly adopting it for massive gains. The technology is helping increase productivity and competitiveness for both organisations and employees.

In a fireside chat at YourStory’s India Tech Leaders’ Conclave 2024 held in Bengaluru on June 21, 2024, Rajesh Ramdas, Senior Director, Field Engineering, Databricks, offered insights into how GenAI is going to revolutionise the innovation landscape. The conversation was moderated by Ipsita Basu, Director Creative Content, YourStory Media.

The age of GenAI

Ramdas pointed out that GenAI has taken the world by storm. Speaking about his work at Databricks, he highlighted that almost 90% of the organisations he has worked with are either invested or experimenting with the new-age technology. Moreover, 75% of the CEOs who lead these organisations believe that GenAI will give them a competitive edge.

“Interestingly, employees who have started using GenAI have seen 40% productivity gains. This is at the organisational level. If we speak at the model level, today we have five to six models, most of which are in open source. So one can witness the evolution of many models that are helping multiple organisations,” Ramdas shared.

While these models are proficient in general knowledge, customers need help in answering specific questions. Experimentation with GenAI was happening but many organisations are not taking it to the production level.

“One of the biggest concerns is around AI regulations. The second is data privacy. Finally, the cost is an important consideration. Is it worth running it at the production scale?” he said.

The challenge conundrum

The challenges with GenAI can be categorised into three levels. The first challenge is at an organisational level, where there's a discussion on who owns the model as well as the person responsible for it.

“The second challenge is about data privacy (not only security of data but of the entire estate). And finally, how do you ensure you are able to address cyber attacks that are happening on the AI models?” Ramdas said.

There are also concerns at the model level. When models come in, they do well from a general knowledge perspective. But if specific questions are asked around churn or pipeline, the technology is unable to answer.

Safety is also a significant consideration, Ramdas said. Wrong answers, if generated, must be caught before they go to the customer.

“When you look at any application that is using GenAI, it has to look at the end-to-end quality and not just at a specific answer. Second, the observability around model drifts is important,” he added.

As far as solutions go, there are multiple things that must be taken into account to address concerns around GenAI. Certain business rules must be laid down to ensure people don't misuse the technology.

“We are also seeing the concept of smaller models. The final aspect is about data. What's happening with us is that data is fragmented. It is equally important to understand the semantics of data in English language. Having governance on top is important to ensure success,” Ramdas shared.

Best practices

Most organisations with a competitive edge in the industry, including the likes of Uber and Netflix, have ensured their data is relevant.

“It's essential to have a single focus. Can you have one set of data and not multiple sets of data? Another best practice is to have governance at the data layer. Having governance in different layers is difficult to manage, so bringing it from the applications to the data itself makes it easier to address,” explained Ramdas, adding that controlling data leakage is crucial.

It's also important to ensure that the model quality output is monitored and corrected at appropriate times. Having the right set of tools as well as processes can help address the problem.

What’s in the future?

When ChatGPT first came in, it was asked to write emails on certain topics. Currently, people are using it to generate images and videos.

While this is progress, the challenge lies in the technology looking for information on the Internet, which might lead to copyright issues.

“One of the customers we work with is Shutterstock and they have a lot of images. They created this model and made it available to their customer. When someone uses their model to create different images, they know their images are based on their own data and they don't have to worry about AI and legal risks,” Ramdas said.

In the future, the focus should be on making anything that's non-English easy to consume. “GenAI will eventually not just give you answers but it will also take action,” Ramdas concluded.