Korea’s Naver joins generative AI race with HyperCLOVA X large language model

Another big gun is entering the AI race.

Korean internet giant Naver today announced the launch of HyperCLOVA X, its next-generation large language model (LLM) that delivers conversational AI experiences through a question-answering chatbot called CLOVA X.

The company said it has opened beta testing for CLOVA X in English and Korean and will make HyperCLOVA X available to enterprise users, allowing them to customize the model on their own data. It also plans to add an AI function called Cue into its search engine, much like what Microsoft has done with Bing, by November 2023.

The move comes at a time when companies across sectors are racing to build AI into their internal workflows to drive efficiencies while the vendors providing these services are going all-in on new capabilities to make the implementation easier. Just recently, OpenAI, which started the generative AI wave, announced that its GPT-3.5 Turbo model can now be fine-tuned on enterprise datasets, while Midjourney expanded its tool with a new generative infill feature.

What to expect from HyperCLOVA X?

HyperCLOVA X builds on its predecessor (HyperCLOVA), which has more than 204 billion parameters. Naver hasn’t shared the exact number of parameters for the new model but it does note that it has learned 6,500 times more Korean data than OpenAI’s ChatGPT (powered by GPT-3). This makes the model, and CLOVA X, particularly useful for localized experiences where it can understand not only natural Korean-language expressions but also laws, institutions and cultural context relevant to Korean society to provide answers.

“HyperCLOVA X … improves on the previous LLMs, which are prone to giving hallucinations and wrong information, and have shortcomings in providing up-to-date information or performing calculations,” the company writes on its website. 

Naver also plans to make the technology multimodal, so that HyperCLOVA X will be able to generate not only text outputs but images, videos and sounds. For example, users could edit photos “just by attaching a file and chatting,” the company explained in a blog post, while noting that the functionality would be added at a later stage.

Integration with enterprise systems

In addition to general use via CLOVA X, Naver’s new generative model will be open to customization by global enterprises. This will transform the generalist model into a specialist one, allowing teams to use it in their desired workflows, much like the way OpenAI provides its GPT family of models. 

“You can tune HyperCLOVA X in the direction you want by using the data required by each industry group. Depending on the field you work in, there are endless possibilities such as ‘HyperCLOVA X customer service,’ ‘HyperCLOVA X coding,’ ‘HyperCLOVA X home appliances,’ etc,” the company writes.

In the area of ​​customer service, Naver explains, the model could automatically classify and analyze customer inquiries to help agents plan scenarios for dealing with customers. In marketing, it will be able to create marketing phrases tailored to the characteristics of the company or provide a summary of marketing reports.

The race is on

With the launch of HyperCLOVA X, Naver is moving to take on other leading players in the generative AI race. These include ultra-scale providers such as Google and Microsoft-backed OpenAI as well as niche vendors like Midjourney.

The company has a 500-strong AI team and is working with Samsung Electronics to build an optimized AI semiconductor solution, which is to be one-tenth the size of its existing one and offer more than four times the efficiency.

According to McKinsey, generative AI could add $2.6 to $4.4 trillion annually to the global economy. That’s far more than many countries’ current GDP.