In a groundbreaking development for the South Slavic language-speaking region, Aleksa Gordic, the founder of Runa AI, has unveiled YugoGPT, a robust generative language model tailored specifically for Serbian, Croatian, Bosnian, and Montenegrin (BCMS).
YugoGPT touted as the largest generative language model for these languages, aims to offer a versatile AI assistant capable of understanding text, answering questions, and addressing the unique needs of individuals and businesses in the BCMS-speaking nations.
Empowering local language needs
YugoGPT emerges as a response to the limitations faced by users outside the English Natural Language Processing (NLP) sphere. While English-centric AI models like ChatGPT exist, they often lack the context and relevance required for non-English-speaking regions.
Gordic emphasizes the advantage of YugoGPT, stating that it allows for training a large language model (LLM) that caters specifically to local requirements.
For instance, Gordic points out that asking ChatGPT about tax-related queries typically assumes a U.S. context, potentially leading to inaccuracies or irrelevant information for individuals residing in Southeastern Europe.
With YugoGPT, users can anticipate more accurate and contextually relevant responses, aligning with the intricacies of the BCMS languages and local practices.
A multilingual vision
Aleksa Gordic, the mind behind this innovative venture, firmly believes that artificial intelligence should serve the needs of every language community. Runa AI is committed to building multilingual foundational models to empower enterprises and individuals across diverse linguistic landscapes.
While YugoGPT represents a significant leap forward in addressing the specific needs of BCMS languages, Gordic envisions expanding its reach and capabilities further.
To accelerate development, Runa AI is poised to raise seed funding, acquire a GPU cluster, and establish an enterprise-level LLM platform.
Unique capabilities of YugoGPT
YugoGPT’s capabilities are not to be underestimated. Gordic proudly claims that YugoGPT 7B surpasses competitors like Mistral and LLaMA 2 from Meta (formerly Facebook), officially establishing itself as the premier open-source LLM for Serbian and other HBS languages.
What sets YugoGPT apart is its commitment to providing access to model parameters, offering individuals and companies unmatched flexibility for customization in various fields such as finance, taxes, psychology, and more.
Addressing data security and privacy concerns
One prominent concern in the age of AI is data security and privacy. Many companies are increasingly wary of relying on American APIs that require sending sensitive data to third-party servers. Gordic recognizes this trend and highlights a growing preference for deploying AI capabilities directly on in-house computing systems.
By granting access to model parameters, YugoGPT puts organizations in control of their AI functionalities. This increased control mitigates concerns associated with third-party involvement, ensuring data security and privacy while benefiting from state-of-the-art AI capabilities.
A journey from Microsoft to deepmind
A strong technology and AI development background marks Aleksa Gordic’s journey to creating YugoGPT. A 2017 graduate in electronics from the Faculty of Electrical Engineering in Belgrade, he honed his skills as a software and machine learning engineer at Microsoft’s Development Center in Serbia, where he contributed to the HoloLens project from 2018 to 2021.
Gordic’s expertise and passion for language models with image and video comprehension capabilities eventually led him to Google’s DeepMind. There, he further specialized in pushing the boundaries of AI, culminating in the creation of YugoGPT, a groundbreaking model tailored to the unique linguistic and cultural context of Southeastern Europe.
