Kim Il-du, CEO of Kakao Brain, announced on the 20th that they will unveil the world's largest-scale multi-modality artificial intelligence (AI) next year.
At a press conference that morning, Kim explained, "We plan to build the world's largest-scale image-text multi-modal dataset and release it along with the model." Multi-modality refers to AI models that operate across text and image learning. Multi-modal AI can simultaneously understand various domains such as text, images, and videos, producing complex outputs.
He added, "Google has built a dataset of 1.8 billion entries, and OpenAI has created 1 billion. Kakao Brain has constructed a dataset of 2 billion entries and will partially release a model capable of understanding images and text simultaneously in January next year."
Kakao Brain previously unveiled the super-large AI models ‘KoGPT’ and ‘minDALL-E’. KoGPT is a Korean super-large language model. minDALL-E is an image generation model that creates desired images in real time when users input text commands. It was pre-trained on a dataset of 14 million text-image pairs and has 1.3 billion parameters.
Kakao Brain has identified 'education' and 'healthcare' as fields where its AI will be applied. He said, "We believe that challenging issues in areas such as education and healthcare can be solved." Kakao Brain also plans to continue collaborating with Kakao’s in-house independent healthcare company (CIC). He stated, "Kakao Brain will lead differentiated research technologies and ensure these technologies are applied to important real-world problems. We are researching to realize human-like AI based on an image-text dataset of around 2 billion entries, and we are also considering ways to contribute to the overall AI industry by partially releasing this."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
