THE TECHNOLOGY INTERVIEW
requiring substantial computing resources, IBM has pursued a different path with its open source Granite model family which targets specific business use cases.
“ Granite is a family of small, agile, specialised AI large language models. They’ re not the 750 billion parameter models out there,” Leon explains.“ You can run them on the cloud or on premise with a smaller number of GPUs.”
This approach delivers practical advantages in cost and efficiency.“ Our models are two-three times more cost-effective and run more efficiently,” he claims.
The transparency of these models addresses growing concerns about data usage and training sources affecting larger, proprietary systems.“ The big issue with proprietary‘ black box’ models is you don’ t know what they’ re trained on. We pivoted early towards transparency,” Leon says.
Lessons from field deployments point to targeted approaches Drawing from IBM’ s experience implementing AI across industries, Leon identifies several patterns separating successful projects from failed experiments. The central factor, he argues, is focusing on specific use cases with measurable outcomes.
“ It’ s all about the use case and quantifiable outcomes. I’ ve seen huge numbers of pilots that went nowhere,” he observes.“ Where we pivoted quickly was ensuring a use
technologymagazine. com 49