Select Star (CEO Se -yeop Kim) is the primary domestic company to be accountable for the worldwide -scale Red Team Challenge. Through this, the goal is to develop into a representative artificial intelligence (AI) reliability and safety verification company.
Hwang Min -young said that Select Star will probably be accountable for the Red Team Challenge, which will probably be held on the Barcelona Mobile World Congress (MWC) on March 5.
This event is a worldwide scale red team event of the World Mobile Telecommunications Business Association (GSMA) organized by MWC. Greater than 200 people world wide will participate, including the LG Uplus team in Korea.
The Red Team Challenge is a technique of inducing potential threats or harmful products to confirm the protection of the AI model. This permits you to check data bias or misuse.
Hwang Min -young said, “Based on the know -how obtained on the domestic event, we develop the AI platform essential for this event, ask questions and evaluate the response.”
Select Star has participated as a representative partner within the domestic AI Red Team Challenge, organized by the Ministry of Science and ICT in April last yr, and has held a big -scale event with greater than 600 participants. Hwang Min -young, who participated as a speaker on the Global AI Telco Conference in Japan, met with the GSMA Asia -Pacific branch manager to introduce the red team challenge experience.
Hwang said, “We didn’t end with a single event this yr, but we discussed the scale of the Red Team Challenge from next yr.” “We expect to introduce Select Star to the European market as a worldwide red team challenge, and to get business opportunities.”
Following MWC, the corporate plans to proceed to pioneer in the worldwide market, including participating within the GTC in San Francisco, USA on March 16-21.
Select Star is an organization that began with data labeling. Nevertheless, it has been diversifying its business for a couple of years, and has released various business models.
Particularly, as you possibly can see within the red team challenge progression, recently focused on AI reliability and stability verification.
In actual fact, on the twentieth, the reliability verification automation solution of the generated AI service was launched. This can be a solution that introduces a red team evaluation method to create a matter that induces harmful answers to AI, and evaluates the AI’s answers to enrich the parts that generally is a problem.
Particularly, it’s explained that it may well increase more efficiency than conventional evaluation methods by routinely generating and evaluating greater than 1 million evaluation questions. He has already secured quite a few corporate users, including the financial sector.
Last yr, the identical context is the introduction of a security filter that removes harmful words to the kids’s chatbot service ‘Momo Jam’. Kid’s chatbot services with Pororo and others must filter out harmful words or expressions more strictly.
For that reason, Momojam has develop into a favourite service for corporations. Tous Les Jours and Momo Jam Collaboration are used for events that create a birthday message for youngsters. “We are going to proceed to proceed cake events using other popular characters equivalent to Hello Carbot due to higher consumer response than expected,” he added.
AI reliability evaluation is extremely demanded in financial corporations which can be very sensitive to the protection of the AI chatbot utilized by the ultimate consumer. Particularly, this requires greater than the performance evaluation of the AI model.
Hwang said, “To dam AI answers that may damage customers prematurely, it’s beginning to define AI performance and safety standards and create questions that may confirm it.” On this process, the core is self -evaluation data It’s to construct three, which can be the principal specialty of a knowledge specialist.
As well as, some chatbots have learned to avoid answers to harmful questions, so that they don’t answer all of the questions of the user, so the user satisfaction could also be poor. On this case, it’s identified that it is crucial not only to avoid harmful answers, but additionally to balance performance and safety, equivalent to inducing other questions or adjusting the extent of answers.
“In some cases, the prevailing model optimization (LLM Ops) solutions include evaluation functions, but most of them focused on easy performance evaluation,” he said. ”Emphasized.
As well as, it’s becoming increasingly more complicated to satisfy each safety and performance because the technology advancement. For instance, most LLM -based apps recently launched are evolving from search augmented generation (RAG) to multi -agent systems. Within the means of combining multiple agents, it’s essential to seek out a particular improvement that may discover and solve the issue.
Because of this, Select Star also said that it’s specializing in the advancement of the answer at the identical time.
Specifically, the ‘Positive-Grained Evaluation’ function that may analyze AI answers and ‘real-time evaluation’ function that will be evaluated for real-time evaluation and performance improvement to enhance performance I’m specializing in doing it.
Hwang said, “Select Star has been developing various business models within the meantime, adapting to the expansion of the AI market.” “We are going to launch the worldwide market on this field with the launch of the Red Team Challenge and AI reliability verification solution.”
By Park Soo -bin, sbin08@aitimes.com
