[하이퍼클로바x] 'Chat GPT 6500 times learning Korean'... Naver to unveil HyperClova X in July 네이버가 오는 7월에 초대규모 AI ‘하이퍼클로바X’를 공개한다. 챗GPT보다 한국어를 6500배 더 많이 학습했다. 네이버는 누구나 자신이 보유한 데이터를 초대규모 AI에서 활용할 수 있도록 지원
[하이퍼클로바x] 'Chat GPT 6500 times learning Korean'... Naver to unveil HyperClova X in July 네이버가 오는 7월에 초대규모 AI ‘하이퍼클로바X’를 공개한다. 챗GPT보다 한국어를 6500배 더 많이 학습했다. 네이버는 누구나 자신이 보유한 데이터를 초대규모 AI에서 활용할 수 있도록 지원
운영자| 입력 : 2023/03/03 [14:55]
공유하기 더보기
'Chat GPT 6500 times learning Korean'... Naver to unveil HyperClova X in July
search-specific 'Search GPT' Development of AI semiconductor solutions with Samsung
Naver will release 'Hyper Clover X', a super-scale artificial intelligence (AI) that learns Korean 6,500 times more than 'ChatGPT' in July.
Naver plans to introduce a 'Search GPT' (photo) service specialized in Korean through HyperClova X. Naver, along with Samsung Electronics, is also making an AI semiconductor solution that is one-tenth the size of the existing graphics processing unit (GPU) and more than four times more power efficient.
Naver announced its plan to release HyperClova X based on these contents at the developer conference 'DEVIEW 2023' held at the COEX Grand Ballroom in Gangnam-gu, Seoul on the 27th. Kim Yu-won, CEO of Naver Cloud, said in a keynote speech, “HyperClova X is an upgraded super-scale AI that combines customers’ own data with HyperClova to provide immediate responses tailored to user needs.” Everyone, even at the national level, will be able to create AI products optimized for their own purposes, creating greater value and creating new business opportunities.” It is interpreted with the intention of creating a 'Korean-style chat GPT' that is specialized in Korean and embodies reliability and up-to-dateness.
Naver's next-generation search technology project 'Search GPT' was also presented in detail. Utilizing OCEAN, a search large-scale language model (Search LLM) specialized for Naver search using HyperClova △Accurate and reliable information through up-to-dateness, repeated verification, and user feedback △Naver such as shopping, pay, and maps Through organic linkage with the service, it provides information optimized for the user's search intentions, such as images, videos, and voices. Kim Yong-beom, chief scientist at Naver Search US, said, "Through the 'Search GPT Project', we can provide a more differentiated search experience to users."
Naver also announced plans to develop semiconductors specialized for AI. Kwak Yong-jae, chief technology officer (CTO) of Naver Cloud, said, “With Samsung Electronics, we have developed a large-scale language model (LLM) that has all the necessary functions for calculation, learning, and inference, while being one-tenth the size and four times more power efficient than existing GPUs. We are making a lightweight AI semiconductor solution,” he explained. “In the case of the data center 'Gak Sejong' to be released this year, we have secured stable power capacity and cooling technology.”
Naver unveils super-giant AI 'Hyper Clover X' in July... Learning Korean 6500 times more than ChatGPT
Enter: 2023-02-27 09:58/Edit: 2023-02-27 15:40
Naver will unveil its ultra-scale AI 'Hyper Clover X' in July. They learned 6500 times more Korean than ChatGPT. Naver plans to support anyone to use their own data in super-scale AI.
On the 27th, Naver held the nation's largest developer conference 'DEVIEW 2023' and announced the release schedule of HyperClova X. Kim Yu-won, CEO of Naver Cloud, who made a keynote presentation, announced that he would unveil a super-scale AI 'Hyper Clover X' in July that can be optimized for customer purposes. CEO Kim said, “HyperClovaX is an upgraded super-scale AI that combines customers’ own data with HyperClova to provide immediate responses tailored to the user’s needs.” We expect to be able to create new business opportunities that will create greater value by creating AI products optimized for
Seong Nak-ho, head of hyperscale AI technology at Naver Cloud, said, “Hyper Clover X is an improved AI that learns 6,500 times more Korean than 'Chat GPT' and can express the AI that users want. If even a small amount of data is combined with the data held by the customer, it is possible to build a super-scale AI product optimized for specific services or areas such as companies. It has been upgraded so that it can immediately provide responses tailored to the user's needs through various interfaces.”
CTO Kwak Yong-jae introduced hyperscale IDC, an AI semiconductor solution essential for ultra-scale AI services, and said, “With Samsung Electronics, we have all the functions necessary for calculation, learning, and inference of LLM (Large-Scale Language Model), while providing the existing graphic processing unit (GPU) ), we are creating a lightweight AI semiconductor solution with a model size one-tenth the size and power efficiency more than 4 times higher than that of )”.
Naver Search US chief scientist Kim Yong-beom explained Naver's next-generation search technology project, 'SearchGPT'. The 'Search GPT Project' utilizes 'OCEAN', a large-scale search language model (Search LLM) specialized for Naver search using HyperClova. He said, “The project is underway with a focus on three things: reliability of information, connectivity with Naver services, and multimodal for effective information provision.”
Reporter Junyeop Kim snoopy@kmib.co.kr GoodNews paper ⓒ Kukmin Ilbo (www.kmib.co.kr) , Unauthorized reproduction, collection and redistribution prohibited
네이버가 오는 7월에 초대규모 AI ‘하이퍼클로바X’를 공개한다. 챗GPT보다 한국어를 6500배 더 많이 학습했다. 네이버는 누구나 자신이 보유한 데이터를 초대규모 AI에서 활용할 수 있도록 지원할 예정이다.
네이버는 27일 국내 최대 규모의 개발자 컨퍼런스 ‘DEVIEW 2023’을 열고 하이퍼클로바X 공개 일정을 밝혔다. 키노트 발표에 나선 김유원 네이버클라우드 대표는 고객 목적에 맞춰 최적화할 수 있는 초대규모 AI ‘하이퍼클로바X’를 7월 중 공개하겠다고 밝혔다. 김 대표는 “하이퍼클로바X는 고객이 자체 보유한 데이터를 하이퍼클로바와 결합해 사용자 필요에 맞는 응답을 즉각 제공할 수 있도록 업그레이드한 초대규모 AI”라며 “개별 서비스부터 특정 기업 또는 국가 단위까지 누구나 저마다 목적에 최적화한 AI 프로덕트를 만들어 더 큰 가치를 창출할 새로운 비즈니스 기회를 만들 수 있을 것으로 기대한다”고 말했다.
성낙호 네이버클라우드 하이퍼스케일AI 기술 총괄은 “하이퍼클로바X는 ‘챗GPT’ 대비 한국어를 6500배 더 많이 학습하고 사용자가 바라는 AI 모습을 발현시킬 수 있도록 개선된 AI이다. 작은 양의 데이터라도 고객이 보유한 데이터와 결합하면 특정 서비스나 기업 등의 영역에 최적화한 초대규모AI 프로덕트 구축이 가능하다. 사용자 필요에 맞는 응답을 다양한 인터페이스로 즉각 제공할 수 있도록 업그레이드했다”고 설명했다.
곽용재 CTO는 초대규모AI 서비스에 필수적인 AI반도체 솔루션, 하이퍼스케일 IDC를 소개하며 “삼성전자와 함께 LLM(대규모언어모델)의 연산·학습·추론에 필요한 기능을 모두 갖추면서도, 기존 그래픽처리장치(GPU) 대비 10분의 1 크기의 모델 사이즈, 4배 이상의 전력 효율성을 갖춘 경량화한 AI반도체 솔루션을 만들고 있다”고 소개했다.
김용범 네이버 서치US 치프 사이언티스트는 네이버의 차세대 검색 기술 프로젝트 ‘서치(Search)GPT’를 설명했다. ‘서치GPT 프로젝트’는 하이퍼클로바를 네이버 검색에 특화한 검색 대규모 언어모델(Search LLM) ‘오션(OCEAN)’을 활용한다. 그는 “정보의 신뢰성(trustworthy), 네이버 서비스와의 연결성(connected), 효과적인 정보 제공을 위한 멀티모달(multimodal) 3가지를 중점으로 프로젝트를 진행 중”이라고 말했다.
"Ready to join the trend of global change" Naver's super-giant AI 'HyperClova X' demo version released In the process of developing an AI semiconductor solution with Samsung to reduce service costs, a model 1/10 the size of existing GPUs and more than 4 times more power efficient The meeting of HyperClova X and Naver Search...'Search GPT' 'Search GPT' Information Source Reliability, Connectivity, Multimodal Emphasis
[Seoul = Newsis] Reporter Oh Dong-hyun = Naver announced on the 27th that it will unveil 'Hyper Clover X', a super-scale artificial intelligence (AI) that learns Korean 6,500 times more than 'Chat GPT', in July.
Naver plans to lead the commercialization of super-scale AI, such as chatGPT's generative AI, so that anyone can use it, and through this, create a domestic super-scale AI ecosystem that is competitive in the global market.
Kim Yu-won, CEO of Naver Cloud, confidently said at the developer conference DEVIEW held by Naver in COEX, Seoul that day, "'Team Naver' is ready to join the global trend of change by concentrating global-level AI technology and capabilities." indicated
“HyperClovaX is a super-scale AI that has been upgraded so that customers can combine their own data with HyperClova to provide immediate responses tailored to user needs,” he said. We expect that we will be able to create new business opportunities that will create greater value by creating AI products that have been developed.”
Since the advent of ChatGPT, interest in super-scale AI services is increasing worldwide.However, linguistic limitations, securing data reliability and up-to-dateness, and cost problems are still difficulties in commercialization.Therefore, it is necessary to understand a specific language and culture well, and to invest in super-scale AI suitable for the industry and environment of the country, and in technology and data for this.
Naver is already proactively investing in its own infrastructure and technology, and working with partners to improve the competitiveness of the domestic super-scale AI industry.
CEO Kim said, "Introducing super-large AI requires astronomical investment and corresponding risks. As a general company, this requires great courage."
"Naver has secured its own technology and infrastructure at a global level through continuous and steady R&D investment over the past 20 years. Naver has business experience and know-how in Korea and the financial sector." I am confident that I am better prepared than anyone else for the changes and shocks that AI will bring.”
Naver recently integrated its technology organizations around Naver Cloud to prepare for the era of super-scale AI.The goal is to make Naver Cloud a super-scale technology company that leads the AI transformation through synergies between super-scale AI technology and cloud/SaaS business.
The key technology for this is HyperClova X.CEO Kim said, “HyperClova is the third largest AI announced in the world in 2021 and the first in Korea. I think we will be able to enjoy our competitiveness."
Following CEO Kim's announcement, Seong Nak-ho, head of hyperscale AI technology at Naver Cloud, emphasized the competitiveness of 'HyperClova X'.
General Manager Seong introduced a demo version of the actual HyperClova X, emphasizing that “for a more specialized and advanced super-scale AI service, it must be learned with data that is close to the user, such as an individual or a company.”
He continued, “As HyperClovaX learns Korean 6,500 times more than ChatGPT and is an improved AI that can express the AI that users want, combining even a small amount of data with the customer’s data can provide a specific service or service. It is possible to build a large-scale AI product optimized for the relevant area, such as a company,” he said.
Super-scale AI requires super-scale infrastructure.Naver is jointly developing an essential AI semiconductor solution with Samsung Electronics to lower service costs.In addition, three data centers specialized in AI, 'Gak', will be opened soon.
In response, Kwak Yong-jae, CTO of Naver Cloud, explained the infrastructure essential for super-scale AI.CTO Kwak introduced AI semiconductor solutions and hyperscale IDC, which are essential for super-scale AI services, and said, "Considering that ultra-large AI will become more advanced in the future, an integrated approach to software-hardware-operating environment is needed." Emphasized.
CTO Kwak said, "With Samsung Electronics, we have developed a lightweight AI semiconductor solution that has all the functions necessary for LLM (Large-Large Language Model) calculation, learning, and inference, while having a model size 1/10 the size of existing GPUs and power efficiency more than 4 times. “Each Sejong of hyperscale data centers to be opened this year has secured stable power capacity and cooling technology to efficiently provide AI infrastructure. With an operating scale four times larger, more efficient and stable operations will be possible.”
In the final session of the keynote, Naver Search US chief scientist Kim Yong-beom explained Naver's next-generation search technology project, 'Search GPT'.
The Search GPT project uses 'OCEAN', a large-scale search language model (Search LLM) that has specialized HyperClova for Naver search, as a backbone.Based on this, Naver models the search flow data of users accumulated over 20 years and guides users to the optimal path to achieve their search objectives.Through this, it is a technology that allows you to better understand search intent and results and generate answers with an emphasis on reliability.
Chief Scientist Kim said, "We are working on a project focusing on three things: reliability of information, connectivity with Naver services, and multi-modal for effective information provision."
"We create information with accuracy and reliability through cross-repetition verification between up-to-dateness and information and user feedback, and provide information optimized for the user's search intention through organic linkage with Naver services such as shopping, pay, and maps." In addition, we want to provide it in an intuitive form by using it with images, videos, and voices," he said. "We expect to provide a more differentiated search experience to users through the Search GPT project."
네이버가 ‘챗GPT’보다 6500배 규모의 한국어를 더 학습한 초대규모의 인공지능(AI) ‘하이퍼클로바X’를 오는 7월 내놓는다.
네이버는 하이퍼클로바X를 통해 한국어에 특화된 ‘서치GPT’(사진) 서비스를 선보일 예정이다. 네이버는 삼성전자와 함께 기존 그래픽처리장치(GPU) 대비 크기가 10분의 1로 줄어들고 전력 효율은 4배 이상인 AI 반도체 솔루션도 만들고 있다.
네이버는 27일 서울 강남구 코엑스 그랜드볼룸에서 열린 개발자 콘퍼런스 ‘데뷰(DEVIEW) 2023’에서 이런 내용을 뼈대로 한 하이퍼클로바X 공개 계획을 밝혔다. 김유원 네이버클라우드 대표는 기조연설을 통해 “하이퍼클로바X는 고객이 자체 보유한 데이터를 하이퍼클로바와 결합해 사용자 니즈에 맞는 응답을 즉각 제공할 수 있도록 업그레이드한 초대규모 AI”라며 “개별 서비스부터 특정 기업 또는 국가 단위까지 누구나 저마다 목적에 최적화된 AI 프로덕트를 만들어 더 큰 가치를 창출하고 새로운 비즈니스 기회를 만들 수 있을 것”이라고 말했다. 한국어에 특화되고, 신뢰성과 최신성을 구현한 ‘한국형 챗GPT’를 만들겠다는 취지로 풀이된다.
네이버의 차세대 검색 기술 프로젝트 ‘서치GPT’도 구체적으로 제시했다. 하이퍼클로바를 네이버 검색에 특화시킨 검색 대규모 언어모델(Search LLM) ‘오션(OCEAN)’을 활용해 △최신성과 반복 검증 및 사용자 피드백을 통해 정확성과 신뢰성을 갖춘 정보 △쇼핑, 페이, 지도 등과 같은 네이버 서비스와의 유기적인 연동을 통해 사용자의 검색 의도에 최적화된 정보 △이미지, 동영상, 음성 등을 제공한다. 김용범 네이버 서치US 치프 사이언티스트는 “‘서치GPT 프로젝트’를 통해 사용자들에게 한층 더 차별화된 검색 경험을 제공할 수 있다”고 말했다.
네이버는 AI에 특화된 반도체 개발 계획도 함께 밝혔다. 곽용재 네이버클라우드 최고기술책임자(CTO)는 “삼성전자와 함께 대규모 언어모델(LLM)의 연산과 학습, 추론에 필요한 기능을 모두 갖추면서도 기존 GPU 대비 10분의 1 크기, 4배 이상의 전력 효율을 갖춘 경량화된 AI 반도체 솔루션을 만들고 있다”며 “올해 공개할 데이터센터 ‘각 세종’의 경우 안정적 전원 용량 확보 및 냉방 기술을 갖췄다”고 설명했다.
