Google’s Korean Bard accelerates local tech companies’ AI race

Home > Business > Tech

Google’s Korean Bard accelerates local tech companies’ AI race

Published: 12 May. 2023, 16:54 Updated: 12 May. 2023, 17:36

Sissie Hsiao, a vice president at Google, announces the availability of Korean and Japanese in Google's Bard artificial intelligence chatbot in a conference on May 10. [SCREEN CAPTURE]

Google’s decision to launch a Korean version of its generative artificial intelligence chatbot, Bard, will likely put local tech companies on edge as they race to develop a Korean-based AI chatbot system.

The Mountain View, California tech giant announced the addition of Korean and Japanese as the first foreign languages to its Bard chatbot and latest PaLM 2 language model at its annual conference on Wednesday.

“Bard is now available in Japanese and Korean, and we’re on track to support 40 languages soon,” said Sissie Hsiao, a vice president at Google.

The announcement of the Korean service, currently available in beta, came as a surprise since Google claims a relatively small market share in the country of 51 million people at around 30 percent.

The Korean version of Bard will pose a direct challenge to similar offerings being developed by domestic players including Naver, Korea’s biggest search engine with the majority of the market share.

The Seongnam, Gyeonggi company will debut HyperClova X, a generative AI tool and SearchGPT, a chatbot based on its AI system, this summer, according to Naver CEO Choi Soo-yeon during a conference call on May 8.

When asked about a more specific release timeframe, a Naver spokesperson said they will be available as early as July.

Asked if the updated Bard would affect the launch of its AI offering, the spokesperson said that “no big change is expected,” although the company is exploring the functions of the Korean-based Bard.

Google and Naver are ramping up their efforts to provide their own generative AI services, following the wild popularity of ChatGPT, a generative AI chatbot developed by OpenAI. Given that the ChatGPT and variants of it tend to display poorer comprehension of Korean than English, Naver has claimed that its service can better cater to Korean users.

Naver said that its HyperClova has 204 billion parameters and its volume of machine-learned data in Korean was 6,500 times larger than GPT-3’s, a language model that powered the initial version of ChatGPT.

Without getting specific about the amount of trained-data, Google CEO Sundar Pichai presented the updates related to the incorporation of Korean.
“Let's say you’re working with a colleague in Seoul and you’re debugging code,” said the CEO during the keynote speech. “You can ask it to fix a bug and help out your teammate by adding comments in Korean to the code. It first recognizes the code is recursive, then suggests a fix. It explains the reasoning behind the fix, and it adds Korean comments like you asked,” he said with actual dialogue threads featured on the projector before him.

With the new updates, Bard has the burden of rebuilding its reputation after a somewhat dreary public debut in February due to factual errors.

Some Korean companies already share the pain. Kakao Brain introduced a Korean GPT-based chatbot ddmm (pronounced “dadaum”) in a test-run but terminated the service a day after because ddmm gave incorrect answers, exposed source codes and took a long time to generate answers.

BY PARK EUN-JEE [park.eunjeee@joongang.co.kr]