National AI Audition Begins: What Are the Strengths of Each Contender?

[Anchor]

The second round of evaluations has begun for the "independent AI model" project, which aims to select a national representative artificial intelligence built with domestic technology.

Reporter Choi Seung-hun tested these AIs to see what strengths each of them holds.

[Reporter]

[(I want to rest today and work out tomorrow.) It's more important to exercise while checking your condition.]

The AI developed by SK Telecom understands human speech and continues the conversation.

If a person interrupts, it stops and listens again, handling workout plans and food orders through conversation.

By processing not only text and images but also voice, it is as if the AI has been equipped with eyes, ears, and a mouth.

[Kim Tae-yun / SK Telecom Foundation Model Division: When both hands are not free, [we designed it] so that you can communicate using voice or video...]

We attached a Korean document on the tax code revision spanning over 300 pages to LG AI Research's chatbot.

When asked to "explain what changes for single-homeowners," it finds the relevant information and organizes the before-and-after changes into a table.

It suggests topics you might be curious about and tailors explanations according to your preferred format.

[Hong Da-sol / LG AI Research Principal Researcher: Let's not make AI feel difficult. [We focused on guiding users] so that as soon as they log in, they think, 'Ah, this is what I need to use.']

To Upstage's AI, we simply input a single sentence: "Make a Minesweeper game."

A playable game is completed in just 27 seconds.

Upon connecting a personal cloud account, it asks what the user needs and even builds a household account book directly.

[Chun Seung-won / Upstage NFM Center Director: We have reached a model where it can automatically execute tasks assigned to it and even produce the final output.]

Motif Technologies joined this evaluation after an additional recruitment process.

A small development team of around 20 people utilized government-provided limited equipment to build the entire AI architecture using proprietary technology.

It performs tasks such as gathering travel information to plan itineraries and writing program code with significantly fewer computations.

[Lim Jung-hwan / Motif Technologies CEO: We are receiving government support funded by everyone's tax money, and we must deliver results that live up to that.]

The second-round evaluation of independent AI models consists of benchmarks measuring performance, expert evaluations, and user evaluations assessing practical usability.

This time, 200 ordinary citizens will also test all four models firsthand starting today (August 8) for four days and score them.

The government plans to narrow down the four teams to three within this month, and after an additional evaluation later within the year, select the final two models.

(Photo: Yonhap News)
(Video reporting: Lee Moo-jin, Yang Hyun-chul | Video editing: Kim Yoon-sung | Design: Kim Ye-ji)
※ Please note: This article was translated by AI and may contain errors.