Bernard Marr's Future of Business & Technology Podcast · Bernard Marr

Why Reasoning Models Are The Next Leap In AI: Inside MBZUAI's K2-Think

·22 min·2 clips

While everyone's building bigger AI models, this university cracked the code on making them smaller but more powerful.

Bernard Marr interviews John Carville, VP of Marketing and Communications at Mohammed Bin Zayed University of Artificial Intelligence (MBZUAI), about reasoning models as the next evolution in AI technology. MBZUAI, approaching its five-year anniversary, was purpose-built as the world's first university dedicated entirely to artificial intelligence education and research. The institution brings together talent from 28 countries and operates research facilities in Abu Dhabi, the United States, and France. Carville, who joined from Meta where he led AI communications, explains how the university takes a uniquely open approach to AI development. Unlike most companies that release only model weights, MBZUAI provides complete transparency by releasing models, weights, data, and source code - eliminating the black box problem common in AI systems. Their latest creation, K2 Think V2, represents a fully sovereign reasoning model built from the ground up. The model focuses on parameter efficiency, delivering frontier-level performance with smaller, more manageable systems that require fewer GPUs and less computing power. This approach makes advanced AI more accessible and cost-effective for businesses and researchers. Reasoning models excel at complex mathematical problems and scientific applications, showing their work step-by-step like students proving math solutions. Key applications include drug discovery, aerospace engineering, oil and gas exploration, and other fields requiring high accuracy and transparency. The model addresses AI hallucination problems through comprehensive self-checking mechanisms. MBZUAI partners with G42 for commercialization while maintaining their open-source philosophy. Looking ahead, Carville identifies world models as the next major frontier in AI development. These systems will move beyond text-based intelligence to physical intelligence, teaching AI to interact with the real world. World models incorporate all modalities - vision, touch, hearing - in a comprehensive system that understands physical environments. Applications could revolutionize robotics, autonomous vehicles, gaming, and entertainment. The technology could dramatically accelerate autonomous vehicle development by simulating every possible traffic scenario without requiring extensive real-world testing. Carville envisions future AI development progressing through world models to eventually include social and philosophical intelligence with human-based value systems. The path to artificial general intelligence remains years away but will advance through iterative improvements. K2 Think V2 is freely available through mobile apps, web interfaces, and direct model downloads from Hugging Face, encouraging community feedback to drive further development.

As heard by us

A clear, practical case for open reasoning models built with efficiency in mind.

Reasoning models are the spine of this PlayNext episode, tying a technical turn in AI to the institutional choices behind building it. John Carville, Vice President of Marketing and Communications at the Mohammed Bin Zayed University of Artificial Intelligence, lays out why…

Read the full review in PlayNext →

Why you'd press play

You want to know whether smaller AI models can reason seriously without demanding a data-center-sized entourage.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Bernard Marr's Future of Business & Technology Podcast