Page 66 - AI for Good-Innovate for Impact

AI for Good-Innovate for Impact

tasks and assistant-style instruction-following capabilities compared to open-source models
and outperforming some mainstream commercial models in non-Latin languages such as Thai,
Khmer, Lao, and Burmese by significant margins.

Compared to existing solutions, which often have strong linguistic bias in high-resource
languages such as English, SeaLLMs is designed to eliminate barriers to accessing cutting-edge
AI technologies for non-English-speaking communities, thereby reducing the risk of cultural
homogenization and preserving linguistic diversity. SeaLLMs offer several benefits, including
improved performance in language understanding and generation tasks for Southeast Asian
languages, respect for local norms and legal stipulations, and cost-effective operation due to
efficient tokenization of non-Latin scripts. However, we also acknowledge potential drawbacks,
such as the complexity and resource intensity of developing and maintaining such specialized
AI models, and the ongoing challenge of balancing model performance across a diverse set
of languages without sacrificing the quality of service in high-resource languages�

Use case Status: The use case is part of a larger research project

Data is publicly available: Yes

Data is privately available: Yes

Repository Link Data repository link

UN Goals:
• SDG 9: Industry, Innovation, and Infrastructure
• SDG 10: Reduced Inequalities
• SDG 17: Partnerships for the Goals

The SeaLLMs project contributes to advancing Sustainable Development Goals (SDGs) 9, 10,
and 17 by fostering innovation, reducing inequality, and encouraging global partnerships.

For Goal 9 (Industry, Innovation, and Infrastructure), SeaLLMs epitomize innovation in natural
language processing by developing language models tailored to Southeast Asian languages.
This innovation enhances digital infrastructure in the region, enabling more inclusive access
to information and technology, and spurring growth in industries reliant on AI and language
technologies.

In relation to Goal 10 (Reduce Inequalities), SeaLLMs directly address linguistic inequalities
by providing high-quality AI capabilities in languages that are typically underrepresented in
digital spaces. By improving AI accessibility and performance for these languages, SeaLLMs
help reduce the digital divide, allowing for more equitable participation in the global digital
economy, and ensuring that technological advancements benefit a wider and more diverse
population.

Regarding Goal 17 (Partnerships for the Goals), the development and implementation of
SeaLLMs likely involve collaboration between multiple stakeholders, including academic
institutions, technology companies, and possibly government entities across Southeast Asia.
This collaborative effort not only exemplifies the spirit of partnership but also ensures that
the benefits of AI and language model advancements are shared widely, contributing to the
collective pursuit of the SDGs.

50