Key pannellists included: 

• Frederik Heda, Senior Data Scientist, BP
• Lourens Walters, Senior Data Scientist, IIRIS Informa 
• Huseyn Gorbani, AI Developer, Rider Levett Bucknall 
• Nadav Eiron, Senior Vice President of Cloud Engineering, Crusoe 
• Eleana Kafeza, Lead Researcher, Technology Innovation Institute

Lourens Walters discussed how AI applications are transforming the way stakeholders interact with annual reports, making mundane documents engaging through interactive elements like videos and quizzes. Frederik Heda highlighted the importance of choosing the right modelling approach based on the specific use case, whether it involves quick response times for customer queries or thorough analysis for regulatory documents. Both emphasised the necessity of evaluating AI models effectively to ensure they meet customer needs and provide business value.

The discussion also covered the evolution of foundation models, with Eleana Kafeza explaining how smaller models are becoming more efficient without sacrificing performance, exemplified by the Falcon H-1 series. Nadav Eiron and Huseyn Gorbani focused on the infrastructure challenges and integration issues, noting that AI development requires robust systems that can manage and secure large datasets across multiple clouds. They stressed the need for abstractions and commonalities in AI infrastructure to facilitate easier deployment and maintenance, which are crucial for widespread adoption and innovation.

Takeaways

Effective AI model evaluation is crucial

Experts stressed that the ability to test and assign good metrics to AI models is essential for developing high-performing solutions. This involves working closely with business stakeholders to obtain accurate ground truth data and using well-defined evaluation methods.

Smaller foundation models are becoming more efficient

Eleana Kafeza discussed the Falcon H-1 series, which integrates transformer and MAM architectures to reduce size while maintaining competitive performance. This trend towards smaller models enables deployment on devices like laptops and phones, broadening AI's accessibility.

The future of AI includes embodied AI and agentic solutions

Panelists predicted that AI would move beyond chatbots to more operational tasks, potentially integrating with robotics. This shift towards embodied AI suggests a future where physical systems interact with humans to provide guidance and assistance in various tasks.