Perplexity AI’s Desi CEO to Zerodha co-founder Nikhil Kamath on AI chatbots: ‘All are same but if used for…’

Perplexity AI’s CEO Aravind Srinivas has said that despite the rapid advancements and intense competition in the generative artificial intelligence (GenAI) space, the core capabilities of leading chatbots currently lack significant distinction. However, speaking on Zerodha co-founder Nikhil Kamath's podcast, he said that there is one area where people may pick the best.
When asked how the Big Tech is doing things differently, Srinivas observed a striking similarity among the major players.
“[The] honest answer right now is all of them are doing similar things. I'll just say it as blunt as it can be. There's not really a genuine differentiation between ChatGPT or Anthropic or Gemini or Grok or Meta AI right now and, of course, Perplexity. You can argue similarly,” he said, attributing this behaviour to the industry's focus on common evaluation metrics.
“Currently, excluding the search part, most models give the same answers,” he explained.
“This is because they are trying to climb on the same benchmarks, same leaderboards. There isn't a lot of qualitative difference,” he explained.

While Perplexity initially differentiated itself by emphasizing accurate sourcing – a feature Srinivas notes competitors are now emulating – he suggested current users' choice often boils down to specific task suitability.
“In the beginning, the differentiation was we were the only ones to make sure you always had sources for everything – highly accurate sources, fast answers and so on but everybody else is also realising that the real values in search even more than free form chat. And they're trying to put sources for almost any response so I think right now we're in this weird phase where like all AI chatbots seem similar,” he added.
However, the one area Srinivas pointed out that makes all chatbots different is fact checking and research.
“If you use AI for fact checks and research, financial research etc, you would pick whichever model does that best,” he commented.
Nevertheless, Srinivas predicts this landscape of similarity is poised for a shift. The crucial differentiator in the near future, he argued, will be the development of “agentic behaviour” – AI systems capable of performing tasks, not just generating text.
“Answering questions will be seen as a commodity—just answering in text will not cut it,” Srinivas made a prediction for 2025. He believes future advancements will involve AI generating richer outputs like graphics and charts and, more significantly, actively “doing things” for the user.
This evolution towards action-oriented AI, Srinivas noted, is dependent on sophisticated reasoning capabilities, which he believes are only just beginning to function successfully within AI systems.