On August 28, language data specialist company Flitto held its first press conference in celebration of its 13th anniversary, unveiling its 'Flitto 2.0' vision and the current state of hyper-personalization technology. Since its establishment in 2012, Flitto has operated a crowdsourced English translation service and platform business. However, as large language models (LLM) like GPT have become a major focus in the AI market, the company has expanded its business scope to become a language big data specialist and real-time translation service provider.
He continued, "Global big tech companies like AWS, Apple, and Google use our services, and they are utilized beyond IT in fields such as healthcare, finance, education, and culture. The trend is expanding translation services from conferences to job training, lectures, YouTube, and other audio and video content." The Core Driver of Flitto's Growth: Live Translation and Hyper-Personalization
Live translation is a feature that interprets spoken content in real-time. Unlike general translation, which takes time to process after input, it translates the content immediately with only a slight time lag. With AI performance improving, any company can quickly implement this. However, accuracy is more important than speed.
The keyword that penetrates this part is hyper-personalization. Flitto's hyper-personalization technology provides real-time translation and correction results tailored to the specific context, tone, and conversation style of the individual as they use the translation more. Existing translation technology could not properly recognize proper nouns or unclear pronunciations, but it corrects content accurately by pre-learning specific data.
CEO Lee stated, "Corporate services currently apply hyper-personalization. For example, even in the medical field, difficult technical terms and pronunciations differ for each person. By pre-entering and learning papers or related materials through hyper-personalization, difficult content is conveyed exactly as it appears in the paper."
Collaboration with the National Institute of Korean Language, Participation in Independent AI Foundation Business
Flitto continues to expand its possibilities in the AI market. It is the only company participating in the National Institute of Korean Language's corpus construction project for five consecutive years and recently signed a memorandum of understanding with domestic AI semiconductor company FuriosaAI to research replacing GPUs in inference tasks. It also participates in the independent AI foundation construction project with the Upstage consortium.
CEO Lee stated, "The data refinement accuracy desired by the National Institute of Korean Language is 99.9%. However, the languages needed for data refinement are difficult and rarely used, such as Khmer and Tagalog. To address this, Flitto enhances data using reinforcement learning with human feedback (RLHF) and chain of thought (CoT), and has obtained TTA data quality certification," adding, "Thanks to these efforts, 75% of Flitto's data sales are overseas, with the largest sales in the U.S. and Japan."
He continued, "The collaboration with FuriosaAI aims to research replacing GPU inference with NPU. Some global companies want to operate Flitto services on their own servers for security purposes, and in such cases, we are exploring ways to utilize NPU instead of GPU," adding, "Currently, we are working only on text, and in the long term, we plan to collaborate to see if more functions, including vision recognition, can replace NVIDIA GPUs." Flitto at 13 Years: Time to Move in a New Direction
Flitto's next step is 'Flitto 2.0.' CEO Lee stated, "We can proudly say we have the highest quality language data. Nonetheless, the next goal is to further enhance and elevate the quality of our data assets. Now, it's time to soar with the new goal of Flitto 2.0 beyond Flitto 1.0. We will be able to provide more global news and growth-related updates in the future."
Starting as a startup, the burden of being the first company to be listed under the special business model listing has always been a challenge for Flitto. However, thanks to the growth of the AI market, Flitto has also entered a rapid growth phase. In March this year, Flitto signed a KRW 11.3 billion language big data supply contract with a global big tech company, and in July, it signed a KRW 6.4 billion contract to supply data for AI-based language model research and development. Based on this, it has recorded three consecutive quarters of profitability. Flitto is now embarking on a broader path from a translation platform company to a language big data company.
IT Donga Nam Si-hyun (sh@itdonga.com)
AI-translated with ChatGPT. Provided as is; original Korean text prevails.
ⓒ dongA.com. All rights reserved. Reproduction, redistribution, or use for AI training prohibited.
