Thailand reduces foreign tech reliance with home-grown AI system

Thailand is taking proactive steps to lessen its dependence on foreign technologies, with a particular focus on artificial intelligence(AI). To this end, the nation is intensifying its efforts around the development of a home-grown version of the generative pre-trained transformer (GPT) model known as Open ThaiGPT, said Chai Wutiwiwatchai, executive director of the National Electronics and Computer Technology Center (Nectec).

“Having our own GPT model will cater to local needs by fostering a robust local ecosystem. It’s a strategic move to reduce risks associated with over-reliance on foreign technologies.”

GPT, a large language model (LLM) technology, is versatile, and capable of a wide array of tasks, from generating original content to summarising text and writing code.

In April 2023, Nectec joined forces with three organisations to kickstart the Open ThaiGPT project. This next-generation AI initiative leverages the Thai language. The collaborating organisations are the Artificial Intelligence Entrepreneur Association of Thailand, the Artificial Intelligence of Thailand (AIAT), and the NSTDA Supercomputer Center.

The project has resulted in a state-of-the-art, customisable AI chat-based assistant that comprehends the Thai language. Its development has been supported by the Digital Economy and Society Ministry and the Higher Education, Science, Research and Innovation Ministry, with the AI chatbots finding use in varied sectors, including tourism and medical, and document search systems for Thai-language R&D papers.

Thepchai Supnithi, director of Nectec’s AI Research Group and vice-president of the AIAT, shared that the Open ThaiGPT project has made significant headway. It has developed a Thai chatbot system with capabilities on par with OpenAI’s ChatGPT. What’s more, it has the flexibility to connect to external systems and retrieve data. It’s also expandable, customisable, and developed as free open-source software.

Open ThaiGPT

The Open ThaiGPT system has grown from 7 billion parameters to 13 billion, and there are plans to expand it further to 70 billion parameters, said Thepchai.

“Like Japan, South Korea, Norway and other European countries, Thailand is well-positioned to have its own language GPT model. Singapore, for instance, is developing Southeast Asia’s first LLM ecosystem, customised to the region’s diverse cultures and languages.”

The Open ThaiGPT platform can serve specific business uses in data management. Businesses can utilise it for tasks such as information searches, and it can serve as a foundation for foreign tech and business firms, including the Government Big Data Institute, to use the framework.

GPT models have potential use in government services and private sectors like banking and industry.

OpenAI’s GPT Store, as reported by Reuters, provides a marketplace for personalised AI applications. Users can build and discover GPTs customised for tasks like teaching math or designing stickers, said Chai.

“A GPT store would offer further revenue opportunities for developers. However, they need to work with linguists to construct extensive databases.”

GPT store

Thailand is considering establishing its own GPT store to offer custom GPT solutions within the country. This would ensure security, privacy, standardisation, and transparency for developers to support specific business uses.

OpenAI’s GPT store could also enable developers to create custom GPTs to reach consumers, offering a new source of revenue.

Chai proposed guidelines on ethics to ensure transparency in LLM data, explainable AI, and no bias. These guidelines would apply to GPT models used for government-related tasks and services, reported Bangkok Post.

Thailand is set to host the 62nd annual meeting of the Association for Computational Linguistics in Bangkok from August 11-16 at CentralWorld. The event will feature an exhibition of global tech firms engaged with the AI movement, Thepchai said.

“The event will expedite AI project development. A two-year project could be reduced to 18 months through collaboration among experts.”