OpenAI’s API access to the public has opened Pandora’s box. People have been rather quick to put it to various uses. Chatbots before OpenAI had limited capability, but chatbots are being transformed with ChatGPT and Metas Llama LLM (large language model) models. One unique application of such a large LLM is Woebot, AI-driven mental health. Woebot uses natural language processing and understanding to support and assist individuals experiencing anxiety, depression, or other mental health concerns.
While the uptake in LLM is over the top among the English-speaking population of the world, there is a perceptible disinterest in the vast Asia Pacific region, which has a diverse language, culture, people, customs and geographical challenges. When we looked at this phenomenon closely, we came up with a few possible answers to this dichotomy.
Language Barriers
The Asia Pacific region is home to many languages and dialects. While LLMs are trained in multiple languages, their proficiency and understanding of specific regional languages or dialects might be limited. This could impact the LLM’s ability to communicate effectively or understand the nuances of local contexts.
Cultural Nuances
The Asia Pacific region is known for its rich cultural diversity. LLM’s training data might not be comprehensive enough to capture each culture’s intricacies and subtleties, leading to potential misunderstandings or insensitivity.
Local Knowledge
LLM’s training data, which only extends until September 2021 (in the case of ChatGPT), may not be up-to-date or inclusive of recent developments and events in the Asia Pacific region. Additionally, the model’s knowledge of specific local customs, regulations, or current affairs might be limited or outdated.
Technological Infrastructure
The effectiveness of AI models like an OpenAI’s LLM depends on the availability and quality of internet connectivity and digital infrastructure. Some areas in the Asia Pacific region might have limited or unreliable connectivity, impacting the model’s ability to provide timely and accurate information.
Bias in Training Data
LLM training data is derived from various sources, which might contain inherent biases. These biases may affect an LLM’s ability to provide unbiased information or analysis. They might limit the model’s effectiveness in understanding and addressing the needs of diverse communities within the Asia Pacific region.
Legal and Regulatory Compliance
Different countries in the Asia Pacific region have varying legal and regulatory frameworks that govern the use of AI technologies. Navigating these complexities might pose challenges in ensuring that an LLM complies with all relevant laws and regulations in the region.
However, it should be noted that LLMs are also gaining traction in the Asia Pacific. For instance, the ability of ChatGPT 3 to converse in Kannada, a language spoken in the state of Karnataka in India (whose capital is Bangalore), was zero. Yet, within months when ChatGPT 3.5 and now 4 arrived, it can understand and converse in Kannada, though with many errors. As time flies by (in months or less) with the release of ChatGPT 5 and onwards, this ability will grow by leaps and bounds. So Asia Pacific’s wait for a complete – as complete as its English language cousin – LLM that can understand its various uniqueness and become a potent tool is not long.
