Keynote: Generative AI for Thai Document OCR
Event: SCBX Unlocking AI EP4, Computer Vision: How AI See Things Like We Do
Collaboration: SCBX and Insiderly.ai
Venue: SCBX NextTech, Siam Paragon, 4th Floor
Speaker: Dr. Kobkrit Wiriyayuthakorn, President, AIEAT and CEO of the company. iApp Technology
Few people like to do paperwork because it is a complicated, complex task that does not strengthen new skills in themselves, but many people should be pleased to know that the use of generative AI makes paperwork that used to be boring. It can be completed in a short time to go to other more useful tasks instead.

Dr. Kobkrit Wiriyayuthakorn, President, AIEAT, briefly talked about the topic of Generative AI for Thai Document OCR at the seminar "SCBX UNLOCKING AI: EP4 Computer Vision: How AI See Things Like We Do" that this document technology is called OCR, which stands for Optical Character Recognition, which is the process of transforming data, whether images or text, in analog format, into digital information that is arranged in an orderly manner.
Dr. Kobkrit He explained that in Thailand, we have been able to use AI to help unpack document data for a long time since 2018, especially by unpacking the data on ID cards through object detection or cutting large pieces of data into small pieces, and then arranging the data in a structured manner to know which parts are which.


But in the future, we will no longer need Object Detection because we can use intelligence like GPT to extract raw OCR text and automatically sort it into structured data.
An additional advantage of not using Object Detection is that we can support documents that do not have fixed templates, such as receipts with various formats, making Thai Document OCR flexible and able to receive new documents immediately.
The advantage of OCR is that it helps workers manage information in documents such as official letters, quotations, receipts, and invoices that can be arranged in a beautiful structure in a quick and orderly manner. No need to waste time typing data one by one and one at a time until the end.




However, there is a disadvantage that the current GPT4 system is still very slow to process data. Especially if the data is entered in Thai. On average, it takes 60-90 seconds to process, which may seem like a short time on the surface, but if you have to handle more than 100 sheets, it means that the cost will increase with it.
The good news is that in Thailand, Thailand's own generative AI called OpenThaiGPT is being developed, which is an AI that collects Thai knowledge and has an important database from the Pantip website.




Recently, the development team tried to take the ONET exam at the Grade 6 level and get a score higher than the average score that Thai children can do. Not only that, but OpenThaiGpt is also good at English. Not only good at Thai.

However, Dr. Kobkrit Admittedly, in the overall picture, our Open AI may not be able to compete with AI in rural areas, but it is a good vision that this will be an important step to help Thai people work better than before, and by the end of this year, OpenThaiGPT version 70b will also be released, which will be many times smarter.
