This is one of the major debates shaking the AI industry at the moment: should we continue to focus on major language models (the architecture behind ChatGPT, Gemini, Claude, etc.) or branch off towards radically different systems? The transition of Yann LeCun, former director of AI research at Meta, to a brand new startup, AMI Labs, is an illustration of this. Rohit Patel, for his part, remained director within the Superintelligence Labs of Mark Zuckerberg’s group where he has worked for 7 years. He believes that major language models constitute a “core of intelligence”a sufficient basis for profoundly transforming the economy and society. The Tribune asked him about this during the EmTech Europe event, organized in Athens by MIT Technology Review and the Greek newspaper Kathimeriní, on March 19 and 20, 2026.
LA TRIBUNE – How do you define “superintelligence”? Some believe that this word is mainly marketing, that it is illusory to imagine a machine surpassing humans in every way… What do you answer them?
ROHIT PATEL – It is unlikely that we will soon reach a situation where AI can do exactly everything that humans can do. Humans can do a lot of things – for example, donate a kidney, which no AI can do. So there are human capabilities that are not even targets for AI. If we imagine a Venn diagram between the capabilities of AI and those of humans, there is a common area: things that humans can do that AI can now do too. But there are also areas where humans outperform them and others where AIs show superhuman abilities, because they possess extremely broad knowledge on many subjects that no human could reasonably accumulate over a lifetime.