韩国“AI大赛”引争议，5款模型中3款“独立性”存疑

Group 1 - The South Korean government's "AI foundational model" competition, launched in June last year, is facing controversy over "technological independence" as three out of five shortlisted teams are reported to have used foreign open-source components, with several models closely resembling Chinese open-source models [1] - Upstage, one of the shortlisted companies, is under scrutiny for allegedly submitting a modified version of a Chinese model, as claimed by Sionic AI's CEO, who pointed out that Upstage's model retained copyright markings from the Chinese model [1] - Other shortlisted companies, including Naver and SK Telecom, are also being closely examined, with Naver's visual and audio encoder being compared to Alibaba and OpenAI products, and SK Telecom's inference code structure being suspected of similarity to DeepSeek [1] Group 2 - Industry insiders have raised concerns that the use of foreign modules as core components contradicts the goal of evaluating domestically developed AI models, questioning the validity of the competition's assessment criteria [2] - Naver and SK Telecom have denied allegations of "shell development," asserting that their core engines are independently developed, while emphasizing the strategic use of verified external technologies [2] - The South Korean government continues to push forward with the competition, with the Minister of Science and ICT calling for a fair and transparent review process, despite the ongoing controversies surrounding the evaluation standards [3]