jasdooit

Naver Cloud and NC AI have been removed from the government’s initial assessment of the “National Representative AI, Artificial Intelligence” program. The “National Representative AI” project seeks to identify key companies that will develop autonomous AI foundational models, with the objective of South Korea emerging as one of the “top three AI powers” along with the United States and China. Although the government initially intended to eliminate just one team from the five, two teams were excluded in this first phase of the competition.

The Ministry of Science and ICT stated on the 15th that “three teams—LG AI Research, SK Telecom, and Upstage—moved to the next phase following a thorough assessment that included 40 points for benchmark (AI model performance), 35 points for expert review, and 25 points for user feedback.” LG AI Research secured the top position in all three evaluation areas: benchmark, expert, and user. NC AI obtained the lowest score in the AI model performance assessment.

Naver Cloud ranked in the top four during an AI performance assessment but was subsequently excluded. Vice Minister Ryu Je-myung from the Ministry of Science and ICT mentioned, “Naver Cloud’s AI model was considered not to satisfy the originality standards, and experts also expressed doubts regarding its originality,” and noted, “it was determined that the model did not conform to the project’s goal of creating a self-reliant AI foundational model.”

Teams that are eliminated get a 10-day window to submit complaints, and the assessment outcomes will be confirmed if no appeals are made.

Earlier, the government intended to remove one team during the initial assessment and then one team every six months, aiming to select a final two teams by the end of this year. However, since two teams were removed in the first round, the government now plans to introduce one more team via a new public hiring process for the second evaluation. Naver Cloud and NC AI, which were eliminated, are eligible to join this additional recruitment. A spokesperson from Naver stated, “There are no intentions to take part in the re-hiring process.”

Some people found Naver’s early elimination surprising, considering its status as a top domestic AI company. Naver Cloud was seen as having advanced technology, frequently mentioned along with LG AI Research as one of the “two major forces in domestic AI.” Even members of the AI industry were taken aback by Naver’s early exit. However, following the disclosure last month that Naver’s AI model utilized a Chinese-developed “vision encoder,” allegations surfaced that it was not entirely built from scratch, prompting speculation that Naver might be among those eliminated.

In contrast to the other four companies that created large language models (LLMs), Naver built an omnimodal model that can interpret images and videos. The vision encoder functions as the “eyes” of the model. AI specialists noted that Naver not only implemented the vision encoder but also repurposed its pre-trained weights. Vice Minister Ryu stated, “An autonomous AI foundational model necessitates setting up weights and retraining them, but Naver utilized existing weights without any changes.”

Industry experts claim that the initial assessment outcomes validated the government’s original goal of creating an autonomous AI foundational model. Should Naver, which utilized a Chinese module, have succeeded, the core objective of the project would have been compromised.

The government further explained the “from scratch” requirements. Vice Minister Ryu mentioned, “AI models need to be created entirely using local technology or open-source software that has no licensing limitations, guaranteeing independence from outside influence or control,” and noted, “even when open-source software is utilized, the weights must be set and retrained.” Open-source software means code that is freely accessible to the public, allowing anyone to alter or share it.