Kakao Brain Introduces One of Korea's Largest Supercomputers, "AI Research Efficiency Improved 100 Times"

[Asia Economy Reporter Buaeri] Kakao Brain announced on the 25th that it has introduced one of the largest deep learning infrastructures in South Korea.

Kakao Brain has adopted the 'Google TPU,' a tensor processing unit developed by Google. The Google TPU is the largest deep learning supercomputing infrastructure in South Korea, surpassing 1 exaFLOPS (100 quadrillion operations per second).

The 'Google TPU' is a large-scale computing infrastructure available on Google Cloud that enables fast data calculation and processing. Instead of installing multiple GPUs (dedicated processors for 3D graphics computation) distributed across several computers, it builds a higher-level system with faster and more complex computations than GPUs within a single supercomputer.

Kakao Brain aims to apply this infrastructure to its Korean language-specialized large-scale AI language model 'KoGPT' to increase AI research efficiency by more than 100 times compared to before.

'KoGPT' is a Korean language-specialized version of OpenAI's language model 'GPT-3,' enhanced for Korean. It performs all language tasks that can be utilized with language, such as determining the positivity or negativity of a given sentence, summarizing long sentences into one line, inferring conclusions from sentences, and understanding context to answer questions.

Kim Il-du, CEO of Kakao Brain, said, "We expect to greatly improve research efficiency by utilizing supercomputers for various large-scale AI models," adding, "We plan to prepare various language models and contribute to building the domestic and international AI ecosystem by open-sourcing them."

This content was produced with the assistance of AI translation services.