At this week’s ABMEC Conference, Guildhawk announced the development of the world’s first domain-specific Small Language Model (SLM) for the mining industry, with the first two modules now in progress. When completed, this ground-breaking SLM will be available within GAI Translate™.
GAI Translate, Guildhawk’s proprietary AI-powered multilingual translation platform, was created as a result of the the first Guildhawk-Sheffield Hallam University Knowledge Transfer Partnership (KTP), back in 2018.
The SLM for Mining is being created to overcome the problems associated with Large Language Models (LLMs) such as hallucinations and document size limits; empower mining clients to win more customers by translating more content for less cost; and improving productivity by removing the repetitive task of having to manually correct results generated by LLMs.
Discussing the need for this model, Jurga Zilinskiene MBE, CEO of Guildhawk, said: “When you use a general LLM to translate documents that contain terms specific to your sector, they often produce errors because they are trained on data that has been polluted. Generic LLMs can generate content that appears credible but is complete fiction and acting on this could result in harm.”
Achieving accurate results from a SLM begins with adopting a disciplined approach to the management of multilingual datasets used to train models. At Guildhawk, this discipline commenced several years ago with the development of GAI’s proprietary Medium Language Model (MLM) and the vast human-verified multilingual datasets. These include domain specific vocabularies built over decades working with global clients.
Alex Shenfield, Professor of Machine Learning, Sheffield Hallam University, said: “High-quality data is the foundation of trustworthy AI. Our work is helping to solve a critical bottleneck in the development of intelligent systems.”
The partnership between Sheffield Hallam University and Guildhawk underscores the commitment of both institutions and the UK government to drive forward technological advancements and apply them to industry-specific challenges.
