Page 67 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

Through these contributions, the SeaLLMs use case illustrates a concrete application of AI
technology that aligns with and advances the objectives of the specified SDGs, demonstrating
the potential of targeted technological innovations to address global challenges.

Future work: If you are given scholarships and resources, what would you propose as future 12-Alibaba DAMO
work on this use case.
• Data collection
• Model development
• Create new variations/extensions to the same use case

12�2�2� Future work

Elaborate proposal:

Given scholarships and resources, the future work on the SeaLLMs use case could expand
across several key areas:
• Data Collection: Enhanced data collection efforts are essential to further improve the
SeaLLMs models. Collecting a more diverse set of high-quality, culturally relevant datasets
across more Southeast Asian languages would help to fine-tune the models for better
accuracy and nuanced understanding.
• Model Development: With additional resources, we could explore the development of
more advanced models or specialized versions of SeaLLMs. This could involve language-
specific models, including more low-resource languages, and scaling the model sizes.
• Create New Variations/Extensions to the Same Use Case: Investigating new variations or
extensions of SeaLLMs could involve exploring multilingual or cross-lingual capabilities
that extend beyond Southeast Asian languages, potentially creating a global language
model that respects regional linguistic idiosyncrasies while facilitating cross-cultural
communication.

12�3� Use case requirements

ITU-T Supplement Y.71 ITU-T Y.3000 series – Use Cases for Specialized Language Models
• SEA-UC02-DESC-001: It is critical that the SeaLLMs model or chatbot built upon the
model can only be interacted via text format, instead of other modalities such as speech
or image.
• SEA-UC02-DESC-002: SeaLLMs was developed to provide strong language processing
and generation capability for local Southeast Asian languages and thus, it may not provide
strong service for other languages.
• SEA-UC02-DESC-003: It is critical that the usage of SeaLLMs must comply with all local
regulations and guidelines related to digital services, data handling, and language use
within Southeast Asian countries.
• SEA-UC02-DESC-004: It is critical that the users of SeaLLMs need to have basic knowledge
and skills of prompting large language models.

51