Naver, AI 'Hyperclova X' New Model Builds... "40% Smaller Than Previous Model, More Powerful Performance"
Naver announced on the 20th that it has built a new artificial intelligence (AI) model, 'Hyperclova X', which is about 40% smaller than the existing model and has more powerful performance. It designed the model with an efficient structure and also showed that operating costs were improved by more than 50% compared to the existing model.
Hyperclova X image [Photo = Naver]
Naver has completed the update of the Hyperclova X flagship model and has begun developing various AI services for users, sellers, creators, business partners, etc. using it.
The updated Hyperclova X model is characterized by its more powerful performance despite its relatively small size, with about 40% of the number of parameters compared to the existing model.
According to Naver, the results of a comprehensive performance comparison using 19 benchmarks for the main learning data, Korean, English, coding, and math, showed that the average scores in all areas surpassed the existing model. In the MMLU (Massive Multitask Language Understanding), a representative benchmark used worldwide, it recorded a correct answer rate of 79.6%, showing a language understanding ability comparable to overseas big tech AI models of similar size.
The multi-modality capability to process various types of data simultaneously has also been improved. The company emphasized that the capabilities of the existing model, which could process not only text but also image data simultaneously, such as visual question-and-answer and understanding of charts and diagrams, have been improved to the performance of the world's top-level model. A new function that can understand not only images but also videos has also been added.
The model was designed with an efficient structure to reduce operating costs. According to Naver, the operating cost of the new AI model has improved by more than 50% compared to the existing model. Naver, which is introducing generative AI technology to major services with large users such as search and commerce (shopping) under the 'On-Service AI' strategy announced last year, is expected to further accelerate the expansion of AI application using this low-cost, high-performance HyperClova X model.
After releasing HyperClova X in August 2023, Naver has been developing the model in line with global AI technology trends. Last April, the company lowered the barrier to adoption of generative AI for companies by launching 'HyperClova X DASH', a lightweight model that can be used for less complex tasks at one-fifth the price of existing models. Then, in August of the same year, the company unveiled the 'HyperClova X Vision' model that can process text and images simultaneously.
Furthermore, it is developing technology that can systematically and comprehensively perform tasks requested by users by enhancing planning and inference capabilities. In the second half of the year, the HyperClova X model that allows natural voice conversation is scheduled to be unveiled.
Seong Nak-ho, Head of Hyperscale AI Technology at Naver Cloud, said, "Recently, technologies that operate high-performance AI models at low cost have been attracting attention, and Naver, which must stably integrate AI into services used by tens of millions of people every day, has also been steadily researching and developing such technologies."
He added, "We expect the new HyperClova X model to become an engine that can provide differentiated AI service experiences to more users, and we will continue to strengthen the capabilities of our main models, such as improving inference capabilities and expanding modalities, and build AI technology that can compete on a global level."
Meanwhile, Naver plans to install this model in its conversational AI service Clova X in March so that users can experience the capabilities of the upgraded HyperClova X more intuitively. It also plans to provide it to corporate customers by launching it through Clova Studio, an AI development tool on the Naver Cloud Platform.
https://www.inews24.com/view/blogger/1815778
Hyperclova X image [Photo = Naver]
Naver has completed the update of the Hyperclova X flagship model and has begun developing various AI services for users, sellers, creators, business partners, etc. using it.
The updated Hyperclova X model is characterized by its more powerful performance despite its relatively small size, with about 40% of the number of parameters compared to the existing model.
According to Naver, the results of a comprehensive performance comparison using 19 benchmarks for the main learning data, Korean, English, coding, and math, showed that the average scores in all areas surpassed the existing model. In the MMLU (Massive Multitask Language Understanding), a representative benchmark used worldwide, it recorded a correct answer rate of 79.6%, showing a language understanding ability comparable to overseas big tech AI models of similar size.
The multi-modality capability to process various types of data simultaneously has also been improved. The company emphasized that the capabilities of the existing model, which could process not only text but also image data simultaneously, such as visual question-and-answer and understanding of charts and diagrams, have been improved to the performance of the world's top-level model. A new function that can understand not only images but also videos has also been added.
The model was designed with an efficient structure to reduce operating costs. According to Naver, the operating cost of the new AI model has improved by more than 50% compared to the existing model. Naver, which is introducing generative AI technology to major services with large users such as search and commerce (shopping) under the 'On-Service AI' strategy announced last year, is expected to further accelerate the expansion of AI application using this low-cost, high-performance HyperClova X model.
After releasing HyperClova X in August 2023, Naver has been developing the model in line with global AI technology trends. Last April, the company lowered the barrier to adoption of generative AI for companies by launching 'HyperClova X DASH', a lightweight model that can be used for less complex tasks at one-fifth the price of existing models. Then, in August of the same year, the company unveiled the 'HyperClova X Vision' model that can process text and images simultaneously.
Furthermore, it is developing technology that can systematically and comprehensively perform tasks requested by users by enhancing planning and inference capabilities. In the second half of the year, the HyperClova X model that allows natural voice conversation is scheduled to be unveiled.
Seong Nak-ho, Head of Hyperscale AI Technology at Naver Cloud, said, "Recently, technologies that operate high-performance AI models at low cost have been attracting attention, and Naver, which must stably integrate AI into services used by tens of millions of people every day, has also been steadily researching and developing such technologies."
He added, "We expect the new HyperClova X model to become an engine that can provide differentiated AI service experiences to more users, and we will continue to strengthen the capabilities of our main models, such as improving inference capabilities and expanding modalities, and build AI technology that can compete on a global level."
Meanwhile, Naver plans to install this model in its conversational AI service Clova X in March so that users can experience the capabilities of the upgraded HyperClova X more intuitively. It also plans to provide it to corporate customers by launching it through Clova Studio, an AI development tool on the Naver Cloud Platform.
https://www.inews24.com/view/blogger/1815778
댓글
댓글 쓰기