In recent discussions surrounding the advancements of artificial intelligence, the emergence of the latest model from ChatGPT, known as GPT-4o, has raised concerns over its potential misuse. This new version, which boasts impressive image generation capabilities, has been exploited by fraudsters to create fake money transfer slips in both English and Thai. The Facebook page kafaak has issued a warning, showcasing examples of these forgeries and emphasizing that "the latest update of ChatGPT allows for the creation of highly realistic bank transfer slips (some characters may be slightly incorrect in Thai, but the English and Arabic numerals are flawless, and even if there are errors, they can be easily corrected)."
In an alarming test conducted by the Thairath Money team, GPT-4o was used to generate a fake transfer slip with the sender named Sam Altman, the co-founder and CEO of OpenAI, and the recipient named Thairath Money. The team noted that while the Thai text contained some anomalies, the English text was so convincingly realistic that it was nearly indistinguishable from an authentic document. To add to the sophistication, the AI was also able to create realistic bank logos and QR codes during the test using examples of slips from various banks in Thailand.
Dr. Piyalitt Ittichaiwong, a co-founder of Preceptor AI, echoed these concerns on Facebook, warning medical professionals, nurses, HR personnel, and merchants that "the new GPT-4o model, which can generate images, can create and edit documents like bank slips, medical certificates, or official documents that are almost indistinguishable from the real thing."
However, the worries extend beyond just fake money transfer slips. The potential for other significant documents, such as medical certificates and government papers, to be forged using the same technology poses a substantial risk of widespread harm. Experts have stressed that falsifying financial documents or official papers is a serious legal offense, with severe penalties if used to deceive others, leading to multiple legal ramifications.
To combat this growing threat, experts recommend several verification methods: checking actual account balances directly through banking apps, enabling SMS or LINE notifications, scanning the QR codes on slips to confirm their authenticity with the bank, and verifying information with the original source of the documents, such as calling hospitals for medical certificates. They caution against trusting documents or images that have not been thoroughly vetted.
Krungthai Bank, amid rising concerns over fake slips, has stepped forward to educate its customers on how to verify the authenticity of transfer slips using the Krungthai NEXT application. The verification process is straightforward: users can access the app, select the "scan" function, choose the transfer slip they wish to check, and verify crucial information such as the sender's name, recipient's name, amount, and transfer date. Additionally, Krungthai Bank encourages customers to register for LINE Krungthai Connext to receive alerts about all account transactions, enhancing security further.
As AI technology continues to evolve, it holds great potential for improving efficiency and fostering innovation. However, it can also be misused by criminals if not accompanied by proper judgment and verification mechanisms.
In a separate but related development, Nayarattana Wongsanphatrachan, President of Siam AI Corporation Limited (SIAM AI), spoke at the AI Revolution 2025 seminar organized by Krungthep Turakij. He emphasized the urgent need for Thailand to develop its own Thai language model, referred to as ThaiLLM, to maintain sovereignty in artificial intelligence and protect cultural identity from the influence of foreign models that may not fully grasp the nuances of Thai society.
Currently, most LLM models in use in Thailand are foreign, primarily from China and the West. While these models are effective, they often lack a deep understanding of Thai culture, including subtleties like local dialects and the distinct flavors of Thai cuisine, which are essential to the national identity. If these nuances are not accurately conveyed through AI models, there is a risk of misrepresentation.
Wongsanphatrachan expressed concern that if users, especially the youth, receive distorted information from AI models, it could adversely affect their beliefs and values in the long term. "AI could become a source of knowledge viewed as more credible than parents or teachers," he warned. Therefore, it is crucial to have AI tools developed by and for Thai people that can filter and verify information according to the standards relevant to Thai society.
ThaiLLM is envisioned as a large language model developed with extensive Thai language data to genuinely understand the context and meaning of the Thai language. It can be applied to various natural language processing tasks such as translation, summarization, or creating conversational chatbots that flow smoothly and are on point in Thai.
There are three main strategies for developing LLM: complete in-house development, which requires substantial resources, particularly large processing centers (AI Factories); utilizing foreign AI models, such as AI DeepSeek from China, which, while potentially more cost-effective than ChatGPT, faces hardware limitations due to the unavailability of Nvidia's H800 and H20 chips in Thailand; and leveraging U.S. models, which often do not prioritize the Thai language, leading to inaccuracies due to the mixing of other languages like Chinese and Vietnamese.
To address these challenges, SIAM AI is collaborating with Nvidia to create a Thai language dataset of at least 15,000 sets to train the model to understand Thai without interference from other languages, marking a foundational step towards true AI sovereignty in Thailand.
The future of AI development is not limited to LLM but is moving towards VLA (Vision-Language-Action) models, which integrate visual understanding, language, and action, leading to Embedded AI that will be increasingly utilized in daily life. Nvidia's new chip, Grace Blackwell 200, can perform tasks up to ten times faster than previous computers, with upcoming models like Grace Blackwell 300 promising even greater speed and efficiency.
Rattana further highlighted the importance of participating in GTC 2025 in San Jose, USA, where discussions on AI trends and the development of AGI (Artificial General Intelligence) are taking place. AGI, which can think and analyze like humans, is becoming more integrated into daily life, necessitating that Thailand develop its own AI technology to keep pace with global innovations while remaining true to its cultural roots. ThaiLLM is not merely a language model but a vital key to achieving "AI sovereignty," ensuring that Thailand does not become overshadowed in an era dominated by AI.
