Microsoft wants enterprises to have their own models. The tech giant believes that the time for POCs is over, and ChatGPT was one of the biggest and most successful POCs. With new techniques like RAG and fine-tuning, the time is perfect for enterprises to move forward.

Rohini Srivathsa, CTO Microsoft India and South Asia, at Cypher, spoke about how the last year has seen speed and agility in the innovation around artificial intelligence. “If I were to talk about technology trends and implementing AI for businesses, I would have probably said the same things. What has now changed is the context,” she explained. 

She addressed the hype around AI and mentioned how this isn’t a new phenomenon at all. “We’ve had AI from analytics, machine learning, deep learning, foundation models, and now generative AI which won’t be the last invention either,” she said. 

AI strategy for businesses

To build a strategy around AI it is essential to understand the term itself. She explained that AI has evolved over the years. “Now, you don’t train different models for different purposes. A large language model is able to solve mathematical problems,” she said, giving an example. 

AI is no more what happens in the backend, it isn’t machine learning models alone but also an accessible user interface. “ChatGPT is old news now but it is a brilliant business model because millions of people are able to interact with the model.”

The models are now trained on so much data now they’ve gained something known as ‘emergent capabilities’. “Yes, the language model works on billions of parameters but what it does look like is reasoning. It takes a text and analyses the information and then succinctly gives it responds to the prompt,” she demonstrated with an example. 

These emergent capabilities or multimodalities of this wonderful technology she says can be harnessed by every business if only they have the right approach from the start. “It isn’t new for companies to adopt AI but only a portion of them are able to scale and the reason is that we’re not thinking of business first.” Dr. Srivathsa explained. 

First she says pick a use-case. Elaborating the most popular ones of AI are content generation, summarisation, code generation, semantic search etc. From here build a niche. “Summarisation can be used in the legal or medical field, for example. There is a major opportunity in these fields which has a dearth of AI tools,” she explained. 

It is essential to be very clear about the business problem AI needs to solve. This problem and solution needs to be measured. “‘What gets measured gets done,’ is a wise rule to go by. And have a portfolio approach,” Dr. Srivathsa said. A portfolio approach means knowing that not one program or product will meet all the business needs, and that the company’s needs and goals might change over time. 

The technology strategy that most companies are looking at falls in two categories, building custom models or fine-tuning existing ones. Gartner’s analysis suggests that over 80 percent of companies are engaged in building pre-trained models. “What determines this is thinking about the company’s data estate,” she explained. According to Dr. Srivathsa, the scale of the organisation, the kind of security and the amount of data are the key deciding factors on the kind of approach a company should consider. 

AI and Microsoft

According to Dr. Rohini Srivathsa, CTO of Microsoft India and South Asia Microsoft is betting heavily on Copilot. They’re integrating Copilot to every possible product. Microsoft has already integrated it to all their applications on the Microsoft Office suite. Further, Bing is evolving to be a more effective search engine. The chatbot is now designed to give out contextual information with links according to Dr. Srivathsa.

Talking about their flagship search engine, “Microsoft Bing is changing how we search for information, by also providing context and being more conversational, users can rely on Bing which also provides indices and sources.”

“This has been a long journey at Microsoft with the company examining existing laws and building policies around security of the users data. Microsoft Research is dedicated to AI responsibility, along with engineering these systems,” she concluded.  

The post Cypher 2023: Microsoft Wants Every Enterprise to Build Their Own ChatGPT appeared first on Analytics India Magazine.