KUALA LUMPUR: The country is taking steps towards developing its own Large Language Model (LLM) in Bahasa Malaysia (BM) through collaborations with universities and local companies, says Datuk Mohammad Yusof Apdal.
The Deputy Science, Technology, and Innovation Minister (Mosti) said the ministry is currently exploring the possibility of building an LLM tailored to the Malay language.
Yusof said one key partner in this initiative is Mesolitica, a local company that has already developed the Malaysia Large Language Model (MaLLaM).
“This model was trained entirely using local datasets and is capable of understanding the nuances of the language and culture more deeply.
“Mimos (a local research and development company), Mosti, and Mesolitica are working together to enhance the capabilities of this model and will involve Dewan Bahasa dan Pustaka (DBP) in the future to enrich the Malay language,” Yusof told the Dewan Rakyat on Monday (Nov 18).
He said Malaysia’s efforts align with a broader Asean initiative to ensure responsible artificial intelligence (AI) development and usage.
In February this year, he said Asean introduced guidelines on AI governance and ethics, providing practical resources for companies seeking to adopt AI within a regional context.
“These guidelines aim to promote alignment across Asean and improve the interoperability of AI frameworks among member nations,” Yusof added.
He said Malaysia has also launched its own AI Governance and Ethics Guidelines, introduced on Sept 20.
These guidelines, he said, are designed to ensure that AI technologies, including LLMs, are developed and implemented responsibly, adhering to local values and cultural principles.
The development of an LLM in Bahasa Malaysia is expected to strengthen Malaysia’s AI ecosystem and reduce dependency on foreign AI technologies, he said.
According to Yusof, LLMs can support decision-making, automate tasks, and facilitate research across various sectors by tailoring global AI knowledge to local language and needs.
However, he said developing an LLM is a resource-intensive process, requiring specialised equipment and often involving cloud-based high-performance computing.
“While services from providers like Microsoft Azure, OpenAI, and Amazon Web Services (AWS) can reduce initial hardware investment, concerns about data security persist, as data stored overseas may risk breaches,” he added.
Yusof highlighted that Mosti, through Mimos, is exploring cost-competitive computing infrastructures that can securely handle proprietary data.
He said collaborations with companies like Phison have demonstrated the potential for reducing costs and enhancing data security.
“Mimos’ aiDAPTIV+ platform allows users to customise and optimise AI models on their own premises, minimising the need for expensive cloud computing centres while ensuring that sensitive data remains secure,” Yusof said.
To drive the development of a local LLM, Yusof said Mosti has engaged in discussions with various government ministries and agencies, including the Health Ministry, Digital Communications Ministry, Higher Education Ministry, and the Prime Minister’s Department.
