“LLMs are probabilistic. They don't at all times know what they don't know,” Sikka defined. “They will sound extraordinarily assured whereas being utterly incorrect. While you begin utilizing them in enterprises, you possibly can't simply allow them to run wild. You must put a leash round AI. You want to know when it's making issues up, when it's unsure, and when it ought to defer to a human. That's what we've been specializing in, making AI reliable and dependable at scale,” he stated.
He stated constructing “reliable and dependable” methods means creating AI fashions that recognise when they're unsure and defer to people in these moments. “In our methods, we design fashions that may specific uncertainty,” he stated.
When requested what he meant by “leash,” Sikka defined, “You may consider it as a security boundary. In our methods, we construct fashions which can be conscious of their very own uncertainty. When the mannequin doesn't know, it says so. That's a really completely different method from most present LLMs, which simply full the following token no matter whether or not it's true or not.”
Infosys quietly backed OpenAI earlier than ChatGPT existed
In a uncommon revelation, Sikka additionally shared that Infosys had been one of many first backers of OpenAI again in 2015, lengthy earlier than synthetic intelligence grew to become a mainstream phenomenon.
“When Sam [Altman] and Elon [Musk] had been beginning OpenAI in 2015, Infosys was one of many first supporters,” he stated. “I used to be fascinated by what they had been making an attempt to do — to create secure and useful synthetic basic intelligence. Again then, there was no ChatGPT, no hype round giant language fashions. It was all research-focused, however we believed it was vital for the world and wished to contribute.”
On the time, the collaboration was largely research-focused, with Infosys contributing to early work round accountable and clear AI methods. “It was about long-term impression, not industrial worth,” Sikka stated, including that these rules are extra related than ever as AI spreads throughout industries.
‘20 individuals will change into 5 individuals plus AI': Automation's subsequent wave
Sikka, who has led international expertise transformations at SAP and Infosys, believes the office will change dramatically within the coming years as firms deploy clever methods.
“It's inevitable. The 20 individuals doing one thing at present will change into 5 individuals plus AI tomorrow,” he stated. “That's not a doomsday assertion — it's an evolution. We've seen this earlier than with automation, with ERP methods, with cloud computing. However AI will make this shift sooner and deeper as a result of it touches cognition, not simply execution.”
He cautioned that firms and international locations should adapt. “Firms that don't work out how one can reskill individuals and design AI+human workflows will lose competitiveness,” he stated. “The problem for international locations like India is to make sure that this transition uplifts individuals reasonably than displacing them.”
On India's IT sector, he added, “The companies mannequin has to alter. It may possibly't simply be ‘individuals doing duties' anymore — it has to change into ‘AI plus individuals delivering outcomes.' That's the following part of digital transformation.”
