AI language models not fully reliable, have a 'truth problem', says Zoho CEO Sridhar Vembu

He said, "AI is helpful with jobs like programming, image-processing, and inspection and can be deployed in parking sensors, but when it comes to other areas of deployment, I believe the rollout of generative AI will certainly be much slower than estimated."

Incidentally, a recent Zoho survey revealed that the company's employees began cutting down their use of Chat GPT and generative AI tools. While 30 percent of the company's employees had reduced their usage of generative AI, 15 percent had stopped using it altogether, Sridhar said.

"My own use of generative AI has reduced drastically because I find that it lectures me," he said in a lighter vein, "Typically, you expect an AI chat tool to give you median conventional wisdom, but when I'm hoping to engage in debate with it, I find that it begins to lecture me."

Recently, news reports claimed that Zoho Corp was working on developing its own language model, with Sridhar fronting the R&D project himself. While confirming the development, he said he was unable to promise that Zoho's LLM would solve existing problems with generative AI: "It would be arrogant of us to claim it will close the chinks that exist with present-day AI tools."

He added, "We can't predict results or timelines of cutting-edge R&D, but I'm personally working to solve the problems faced by AI language models. To me, building an LLM is our way of declaring technological independence."

A few months ago, Zoho Corp's CEO was in the news for his criticism of AI monopolies. Sridhar also wrote to the government on drafting an AI regulatory policy, which he said had received acknowledgement. "It's still too early to discuss regulating AI, but there is work in progress," he added.