Oracle's HeatWave generative AI suite rolls out in Korea

Tech

Published: 02 Jul. 2024, 18:11

Oracle's HeatWave generative AI suite rolls out in Korea

Nipun Agarwal, Oracle’s senior vice president of MySQL database and HeatWave development, introduces the company's new AI bundle known as HeatWave Gen AI to Korean press in an online briefing on Tuesday. [ORACLE]

U.S. tech giant Oracle released its HeatWave Gen AI software suite in Korea last Friday, including various features intended to help businesses more easily build and apply generative AI models within its cloud database.

HeatWave Gen AI, which was first announced last week, includes an industry first in-database large language model (LLM), which customers can use without needing to move their data to a different location.

Another feature, known as automated vector storage, allows computers to efficiently find and use data that has been turned into numbers. The automated process enables clients to easily use AI with their data without needing special skills.

“HeatWave Gen AI provides an integrated, data-based solution, which provides security and simplicity to the applications,” said Nipun Agarwal, Oracle’s senior vice president of MySQL database and HeatWave development, in an online press briefing on Tuesday. “It is fully automated and lowers the cost for building generative AI applications.”

HeatWave Gen AI is also equipped with a chatbot that answers queries from developers using natural language or the database programming language SQL.

Agarwal believes that Oracle’s AI offering, previously known as MySQL HeatWave, will benefit industries with unstructured data, such as banks and medical facilities which possess a variety of information in PDFs or nonelectronic legal formats.

Oracle also claims that HeatWave Gen AI guarantees data privacy as its in-database LLMs are trained on public data.

“We send the results of the vector as an input to the LLM,” Agarwal explained. “There is absolutely no risk of data leakage because the LLMs are not trained on this content: It is just sent as an input, and the LLM does not keep any history or preserve any state of the enterprise content.”

The Korean language is supported through two in-database LLMs — Mistral AI’s Mistral 7B and Meta’s Llama 3 — but the results may not yet be as satisfactory as they are in English, Agarwal said.

“But they are improving very fast,” Agarwal added. “I’m sure in a few months that they will catch up.”

BY LEE JAE-LIM [lee.jaelim@joongang.co.kr]