Alibaba research unit unveils LLM tailored for Southeast Asia as e-commerce giant pushes AI in fast-growth region

Alibaba Group Holding’s research unit Damo Academy has launched an artificial intelligence (AI) large language model tailored for southeast Asian languages, highlighting the company’s ambition to grow markets in the wider region.

The Southeast Asia LLM (SeaLLM) was pre-trained on Vietnamese, Indonesian, Thai, Malay, Khmer, Lao, Tagalog, and Burmese data sets, and has outperformed other open-source models in linguistic and safety tasks, the Alibaba research arm said in a statement on Monday.

This is Alibaba’s first region-specific LLM, with Southeast Asia seen as an important growth market. For example, Alibaba’s Southeast Asia e-commerce platform Lazada is targeting US$100 billion turnover by 2030 with 300 million consumers in the region. Alibaba owns the South China Morning Post.

SeaLLM chat, a fine-tuned chat assistant that comes with the LLM, is designed to help businesses using the LLM to engage with Southeast Asia markets.

Damo’s new LLM signals that Chinese companies are still doubling down on the generative AI wave ushered in by OpenAI’s ChatGPT last year. As of July this year, a total of 130 LLMs had been released by Chinese companies and research institutes, leading to what has been termed as a “tussle of a hundred large models” in the country.

Why China is still several moves behind in the AI chess game started by ChatGPT

Alibaba’s LLM Tongyi Qianwen, launched in April, was ranked fourth globally as of Monday among all models tracked by open-source AI platform Hugging Face, which publishes a real-time ranking covering more than 100 generative AI models worldwide.

Hugging Face’s ranking is based on a variety of elements, including scientific knowledge, multitasking accuracy and common sense reasoning.

Damo said the SeaLLM performs better than other LLMs, such as ChatGPT, in non-Latin language tasks, with the ability to interpret and process text up to 9 times longer for non-Latin languages.

SeaLLM also achieved better results in translating between English and low-resource languages – those with limited data for training conversational AI systems – such as Lao and Khmer.

Bing Lidong, director of the language technology lab at Damo, said the SeaLLM can “embrace the cultural richness of Southeast Asia”, adding that innovation is set to empower communities historically under-represented in the digital realm.

Despite having a hundred or so LLMs, analysts have said that the Chinese AI market still faces challenges, including US chip restrictions and the lack of killer apps, in signing up more users in a rapidly-changing market.

Additional reporting by Coco Feng