엔비디아는 3월 20일부터 24일까지 ‘엔비디아 GTC 2023 콘퍼런스’를 개최했다. 생성형 AI 세션도 준비해 생성형 AI의 국내 스타트업을 초청했다. △클레온 △플라스크 △슈퍼톤 △리콘랩스 등 AI 스타트업들은 최신 비즈니스 트렌드, 생성형 AI와 GPU 가속 컴퓨팅, 기술 동향, 직면 과제와 해결 방안을 논의했다.
▲24th Korea AI Day Flask CEO Lee Jun-ho
650 sessions from Jensen Huang's keynote speech through the 24th
Korea AI Day, application of generative AI such as 3D and voice synthesis
NVIDIA held the GTC 2023 conference and presented a future in which AI and humans interact, centered on generative AI such as ChatGPT, which has recently become a hot topic.
NVIDIA held the 'NVIDIA GTC 2023 Conference' from March 20th to 24th. NVIDIA GTC is a global developer conference for the AI and metaverse era, and was held virtually this year as well as last year.
“For nearly 40 years, Moore’s Law has dominated the computing industry, affecting every industry,” said CEO Jensen Huang in his keynote speech. “Computing is advancing at a very rapid pace, and the driving force behind this is AI.”
GTC 2023 features 650 sessions from technology experts and business leaders around the world on research in AI, HPC, cloud, metaverse, generative AI-accelerated computing, and how AI is transforming the world. In particular, he attracted attention by having a conversation with Jensen Huang and OpenAI co-founder Ilya Sutskever.
In particular, at this GTC 2023, various technology talk sessions that will innovate Korean AI research and business for Korean developers, as well as N&up participants and the generative AI Korean Inception startup showcase session, were unveiled on the 24th (Friday).
■ Introducing promising Korean AI startup technologies NVIDIA hosted a Korean AI Day for GTC 2023 and invited domestic startups in generative AI. AI startups such as Cleon, Flask, Superton, and Recon Labs discussed the latest business trends, generative AI and GPU-accelerated computing, technology trends, challenges faced, and solutions.
Cleon is a generative AI startup in the digital human field. In this session, they presented plans for a wide range of areas, including the external appearance of digital humans, as well as the thoughts and actions of digital humans.
“Based on the vision of revolutionizing communication, Cleon focuses on creating more natural-looking digital humans,” said CEO Jin Seung-hyeok. “Through our collaboration with NVIDIA, we will advance our technologies for model lightweighting and real-time solutions.”
Flask is a generative AI startup in the 3D animation field. CEO Lee Jun-ho introduced the technical difficulties encountered while introducing generative AI to browser-based 3D animation and the solutions applied to the product and service roadmap to overcome these difficulties.
“With NVIDIA’s USD support, we were able to collaborate with the global industry, and we also received a lot of help with Tensor RT through the NVIDIA Inception program,” said CEO Lee Jun-ho. “Flask is actually useful in games and movies, for example, it makes it easy to create prototypes when working on animations, and it also makes it easy to exchange feedback with users.”
Supertone is a generative AI startup in the AI voice synthesis solution field targeting the arts and creative fields. CTO Heo Hoon introduced through a demo the efforts Supertone is making based on the 'NANSY (Neural Analysis and Synthesis)' framework to provide an environment where creators can create and precisely control their voices.
The direction of recent voice research is moving toward demanding a hyper-realistic level of voice quality and expression. CTO Heo Hoon said, “We expect that AI will greatly improve the content enhancement and usher in a content renaissance.”
Recon Labs is a generative AI startup in the field of AI-based 3D modeling technology. In this session, Recon Labs revealed the technology it has accumulated based on NeRF (Neural Radiation Field), which is attracting attention in the graphics and 3D fields. For example, it smoothly expresses the surface treatment of an incomplete 3D model shot in video using AI technology.
CTO Yoon Kyung-won said, “Similar to MRI, it is an AI technology that reconstructs spatial information from 2D images, and can directly implement various objects and types of items such as sofas, food, and dolls.”
CTO Yoon said, “We were able to reduce the overall time and cost by more than 40% by efficiently using NVIDIA’s technology, ‘Instant NGP,’ which accounts for more than 90% of the total training portion.”