Introducing ChatEXAONE : Enterprise AI Agent Boosting Corporate Efficiency with Expert AI Insights

This achievement is the culmination of LG AI Research’s rigorous journey over the past few years. LG AI Research first declared our goal to develop large-scale AI in 2021 and unveiled EXAONE. In 2022, we unveiled a multimodal model and continuously made further developments. In 2023, we also announced EXAONE 2.0 with improved performance, as well as three major service platforms developed based on it.

Now, we are pleased to unveil EXAONE 3.0, our first open-source LLM(Large Language Model) with gloabl top level performance and ChatEXAONE, an enterprise AI agent that was developed based on it. 

Enterprise AI Agent 

ChatEXAONE is an enterprise AI agent that helps enterprise users work more efficiently and provides expert-level insights. Unlike many LLMs,  that focus on general content generation and Q&A, ChatEXAONE is specifically designed to meet the needs of enterprise users. Now, enterprise  users can freely upload internal document data without worrying about security, and they can work more professionally and efficiently based on the insights that ChatEXAONE provides.

Many people asked us the same question about ChatEXAONE from planning and development, to its ultimate unveiling. At LG AI Research, we believe that AI will make life better for humans, and with that thought in mind, we worked on research and development in earnest. During that process, we discovered something important. While there are many AI services that solve everyday problems, there is still no high-level AI that can be used in specialized fields and industrial settings. We started developing ChatEXAONE with the idea that it is essential for corporate users and experts to have an AI that they can trust. 

This version of ChatEXAONE is provided to LG Group employees first. This version is equipped with four key features and six major capabilities to help users innovate their work. Here are the four key features, along with the service screen.

Feature 1. Professional Personalization

There’s only one thing left for the user to do : Start a conversation with ChatEXAONE.

Different people need AI for different reasons. How they should use it in their work is also different, which is why ChatEXAONE remembers the user’s job and work expertise and suggests personalized prompts optimized to help users get the most out of the AI. It doesn’t stop there! It also suggests additional prompts to help users gain deeper knowledge and insights. Users don’t have to think about how to best utilize ChatEXAONE, as ChatEXAONE knows how to best utilize itself. 

Feature 2. Document and Beyond (Document-based QA | Web-based QA)

Users are now free to upload their work documents and enter any questions they want to ask.

ChatEXAONE analyzes a variety of data, from internal documents to external data, and uses it to provide answers to the user. It understands documents written in PPTX, PDF, and CSV. It can interpret not only text but also complex diagrams, graphs, and formulas. It can compare and analyze the content of multiple documents and summarize the key takeaways from a single document. It can also use real-time online data during conversations with the user when necessary, such as key trends, news articles, and specialized reports on specific areas, to help the user expand their insights.  

Feature 3. Multimodality (Speech Understanding | Image Understanding)

Experience of knowledge expansion and innovation in work productivity through multimodality begins now. 

Conversations with ChatEXAONE can consist of various forms of data, and text, voice, or images are all possible. Users can utilize data from various tools they use during their work without converting the format. They can upload data in various formats and get insights quickly.

Feature 4. Expert Modules (Coding Assistance | Data Analysis and Visualization)

Code and SQL query generation can now be done by ChatEXAONE.

ChatEXAONE also offers modules for experts in specific fields, such as code generation and SQL query generation for developers and scientists. Users can leverage these specialized features to work more efficiently and create customized workflows by grouping frequently used features within ChatEXAONE.  Experts in the company can save time by utilizing the module’s features and instead devote more of their personal time to focus on other important tasks. 

Key Capabilities 

ChatEXAONE is equipped with six major capabilities that LG AI Research carefully considered during development. For the major capabilities, we evaluated performance using public benchmarks, and for specific functions, we also utilized our own benchmarks. 

ChatEXAONE provides document-based QA, web-based QA, data analysis and visualization, coding assitance, speech understanding, and image understanding capabilities. For each of the six capabilities, we evaluated their performance and found that they performed at the global top level.

ChatEXAONE is based on a web-based search function, which connects to various web APIs to retrieve relevant results and generate sourced answers. Notably, the search function can be customized to meet the needs and expertise of the company, and it achieved performance that outperforms GPT-4 in the G-eval. ChatEXAONE can generate code in several programming languages, including Python, C++, and Java. This capability has also been shown to outperform Meta’s Llama3 model. You can find details of performance evaluation results in ChatEXAONE Technical Report.

Imagining and Looking Forward to Even Bigger, Brighter Futures

We believe that ChatEXAONE will revolutionize the everyday lives of corporate users. It will allow them to complete highly specialized tasks in an efficient way like never before, and how well they utilize ChatEXAONE will eventually become their competitive edge. LG AI Research continues to challenge ourselves so that more users can innovate their work with ChatEXAONE, and so that we can envision a new future. A detailed description of key features and capabilities, as well as performance evaluation results, can be found in the ChatEXAONE Technical Report at the following link. 

ChatEXAONE Technical Report →

EXAONE 3.0 : Our First Open-Source LLM →

CEO Letter →