“Testworks will give attention to not only verifying the security of artificial intelligence (AI), but additionally developing small language models (SLMs) that may be applied to every industry and developing AI services optimized for specific industries.”
Testworks (CEO Seok-won Yoon) is a representative data specialist company. Starting with data labeling for autonomous driving, it has expanded to 3D and multi-modal data. Over time, it has built datasets for over 100 corporations.
And as we have fun our tenth anniversary next yr, we are actually expanding our business to incorporate direct model development and repair provision. For this purpose, the person we hired on July twenty ninth is Vice President Ahn Se-gi.
He served as the top of the ICT Convergence Headquarters at Korea Trade Information and Communication Network (KTNNET) and oversaw the digital transformation of the national trade network. Afterwards, he planned and led quite a few AI projects as the top of information and data security and external cooperation at MarkAny, a security AI specialist. He’s currently in command of Testworks’ AI business division, QA business division, AI-SW functional safety organization, and recent business development.
Vice President Ahn listed three future business directions for Testworks: ▲AI safety verification ▲SLM development ▲AI service development.
“Previously, we mainly did data verification to make sure AI performance, but now we’ll give attention to creating AI models and developing actual services,” he explained. Specifically, he emphasized that the technological capabilities collected up to now have been optimized for the AI life cycle, from data collection to processing, learning, and distribution.
First, the corporate announced that it’s preparing functional safety verification and quality certification for AI products, including autonomous driving software (SW), to judge the reliability of the corporate’s AI.
“To be able to export products overseas, credible AI SW functions and safety verification are essential,” he said. “A 3rd-party certification agency must confirm not only the standard of service but additionally the reliability of the corporate’s operations and evaluate whether it’s following safety guidelines.”
He also added, “We are going to create a verification mechanism in accordance with industry officials and international standards.” In other words, we plan to develop a separate AI for testing to enhance verification efficiency. The plan is to automate the present process that required a number of time and manpower to confirm the security of AI products.
Testworks has already developed the ‘Tedworks’ platform that may evaluate the performance of semiconductor software in real time. The plan is to introduce AI here to reinforce its functions. It also has experience in product performance testing for several NPU manufacturers.
“Multiple countries world wide will concurrently access a single cloud farm to check NPUs and GPUs,” he said, predicting that engineers working in numerous time zones will give you the option to conduct real-time performance evaluations, increasing efficiency.
Based on the experience of existing data construction projects, we also plan to take a position in the event of enterprise SLM. The idea for that is that “there are lots of cases of high-quality data being created in specialized industries comparable to construction, national defense, and aviation through the National Intelligence Service (NIA) data construction project.”
Recently, through NIA’s ‘Super-large AI Diffusion Ecosystem Creation Project’, ‘Trip Chain’ data is being built to secure movement data of transportation users. “That is data that’s used not just for public transportation users but additionally for personalized transportation services and the smart logistics industry by processing passenger automotive movement data,” he said. “It’s a difficult task, but I’m confident because there are lots of existing cases of information construction.”
As well as, he said there’s a powerful demand for AI development for specific services in areas where securing data is difficult, comparable to constructing data for AI for construction site management in response to the Serious Disaster Punishment Act.
As a representative example of expansion within the service field, an SLM-based AI call center solution was cited.
In this fashion, reflecting corporate needs, we’re developing industry domain-specific functions for the present ‘blackolive’ solution. Blackolive is an information processing management solution that supports 4 stages from corporate data collection to de-identification, automation, and processing.
The important thing point is that Testworks is supporting your complete strategy of constructing AI by introducing AI technology to 4 existing business areas, namely Tedworks, Black Olive, the social platform ‘aiworks’, and the AI data quality tool ‘ADQ’.
Lastly, we didn’t forget to emphasise the importance of information.
Vice President Ahn Se-gi emphasized, “Ultimately, what guarantees AI performance is high-quality data,” and “Testworks’ unique strength is that it could possibly go from data quality verification to model and repair development.”
Reporter Park Soo-bin sbin08@aitimes.com
