Naver's Project Connect X will challenge OpenAI's generative tools

Tech

Published: 21 May. 2024, 18:24

Naver's Project Connect X will challenge OpenAI's generative tools

Naver hosted a seminar to introduce the new enterprise AI platform based on its large language model (LLM) HyperClova X in Gangnam District, southern Seoul, on Tuesday. [NAVER CLOUD]

Naver plans to release a new enterprise AI platform for Korean businesses, incorporating the latest automated browsing and large vision language model (VLM) technologies, to compete with OpenAI and Meta.

The service is based on Naver’s large language model (LLM) HyperClova X and is currently known as Project Connect X. It’s still under development, and Naver employees have utilized it for test runs since last November. Connect X leverages automated browsing to extract information from provided texts in order to handle everyday tasks such as writing emails, generating reports and responding to queries.

“Workers spend about 61 percent of working hours performing sideline tasks in communication instead of focusing on real work,” said Naver Cloud’s Lee So-eun, head of Project Connect X, at a seminar in Gangnam District, southern Seoul, on Tuesday, citing a report from the McKinsey Global Institute.

“Our primary aim was to reduce inefficiencies in information retrieval, so we developed a search system and various assistant functionalities to gather scattered information from various internal systems,” Lee continued, adding that the company had experienced challenges during the internal trial process of its enterprise search solution due to its outdated data and data retrieval systems.

However, Lee said, “we observed a high demand in the market for retrieval-augmented generation [RAG] solutions, which can be relatively easily constructed based on refined data.”

But clients' needs, including internal data collection and the drafting of financial statements and contracts, turned out to surpass RAG's current capabilities.

“That is why we are enhancing Connect X through automated browsing, which can upgrade the platform’s capability to respond, and the scope of data understood by RAG has been broadened through a transformation model for unstructured data.”

Both automated browsing and RAG enhance LLM capabilities by incorporating external information. But while RAG leverages a pre-indexed knowledge database, automated browsing can retrieve current data from the web.

VLMs models can comprehend text based on visual inputs such as image captioning or image-text retrieval, which Naver believes can be utilized to understand various data formats in an enterprise context.

Naver plans to release the platform in modular units, which clients can customize according to their needs.

A release date for the platform has yet to be announced.

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