In the latest Intel On AI podcast, host Ryan Carson interviews Ian Gilmour, founder of Euler Digital, creator of Bezoku, member of the Intel® Liftoff program.
In this episode, they discuss Bezoku, a platform developing efficient low-resource language models. Ian highlights the benefits of smaller models, such as reduced power consumption and costs, while emphasizing the importance of cultural representation in language processing.
With support from Intel, Bezoku aims to industrialize workflows and enhance accessibility across diverse communities.
Gilmour describes the challenges he faced in finding tools that could effectively support low-resource languages, such as Haitian Creole.
Building AI for Low- Resource Languages: Bezoku's Innovative Approach
By leveraging a blend of open-source resources, syntactic data, and innovations like reinforcement learning, Euler Digital has been able to create a model tailored for specific dialects. This approach also enables more accurate and inclusive applications, such as polling services that need language adaptability to reach diverse communities.
Bezoku’s development process has involved unique techniques, including the use of “annealing,” where the highest-quality training data is reserved for later stages to maximize model performance.
Gilmour highlights that Bezoku is building its models from scratch, rather than fine-tuning existing large language models, a unique choice that brings specific challenges, but allows for deep customization.
This focus on dialects and underserved languages enables businesses to create more culturally aware, adaptable AI solutions, in fields like customer support and data collection.
The episode provides valuable insights into the complex, multi-disciplinary efforts required to bring AI support to low-resource languages. Euler Digital’s approach showcases the importance of language inclusivity and paves the way for a future where AI can better serve diverse global communities.
Bezoku is set to make significant strides in the field of low-resource language models, enhancing communication and cultural representation. Their commitment to creating efficient, scalable solutions positions them as a key player in addressing the growing demand for hyper-personalized language applications.
As a proud member of the Intel® Liftoff program, Bezoku is well-equipped to refine their technology and broaden their impact.
Stay tuned for their developments as they work to redefine language processing and accessibility for diverse communities. For more information, visit Bezoku's website and follow their journey.
Watch the discussion here: https://www.youtube.com/watch?v=-u5sk5YweuE
As a Software Tools Ecosystem Specialist at Intel, I’ve had the privilege of working on the dynamic GenAI initiative. My focus is on driving engagement with the software developer audience. I'm a proud team member of the Intel® Liftoff for Startups and the Developer Engagement, Relations, and Studio team.