EXAONE Lab, Dreaming of Super-Giant AI to Advance the World

Super-giant AI technology, which has become the core of the future industry, is being researched and developed by a large number of tech companies. LG AI Research has also newly established the EXAONE Lab, further accelerating the development of technology for the advancement of LG’s super-giant AI model “EXAONE.” We’ll be hearing from Jinsik Lee, the leader of EXAONE Lab, about EXAONE’s research direction and vision.

EXAONE Lab, led by NLP (Natural Language Processing) expert

At school, Jinsik focused on the field of natural language processing for speech synthesis, and after graduating, studied the field of language understanding, which identifies human intentions, and the field of conversation using it. He was drawn to the field of language among other areas of AI research because he believed in language being a living thing that changes in response to its surroundings.

Q. What do you think are the characteristics of the language field among AI fields?

“The most familiar form of communication used by humans is language, which is a living thing that evolves over time. For instance, some words from the past may eventually become obsolete, while new words are continuously being created. Another characteristic is that language reflects culture. The expression of the same object or event may vary even within the same language system due to cultural differences. I think this is what differentiates language from other fields.

As most of the knowledge in the world is processed in the form of language, if AI gains the ability to understand language, it’ll be able to acquire an abundance of knowledge from the world around it. What’s more, I believe that this acquired knowledge will be of great help in solving various problems in the real world.”

 

 

Recently, he has been focusing on the commercialization potential of language AI models. Meanwhile, language AI technology has been heavily focused on research and development based on a scientific curiosity about how far the language model can demonstrate its capabilities. However, recently, it is being said that it is a noteworthy technology trend that attempts to increase user contact by applying user-friendly interfaces such as conversation and search to language models, and attempts to continuously improve service quality by receiving feedback from users. However, it has been pointed out that there are still issues that need to be solved technology-wise for commercialization, as there is the issue of “hallucination,” which is an issue with language models-- in other words, these models sometimes generate untrue or biased answers.

 

LG AI Research sees the future of super-giant AI technology

 

 

Before joining LG AI Research, Jinsik was involved in developing voice assistant services, smart speakers, and chatbot services using language AI research knowledge. Then, after hearing that LG AI Research was busy developing attractive super-giant AI technology, he joined as a researcher.

Q. What do you think are the strengths of LG AI Research?

“One of the strengths of LG AI Research is that it’s composed of researchers who are competent, full of ambition, and have the confidence that they can do anything. In particular, there’s a high level of trust among the researchers and a strong sense of camaraderie, so what stands out is their strong will to work together to solve any problem.”

There were very high expectations for the potential and possibilities offered by giant AI, but concerns were raised about areas that could be utilized right away. To learn and operate a super-giant AI model, it is necessary to invest resources such as large-scale infrastructure and data, but it is difficult to find a business model that could generate an appropriate level of profit. He explained that this concern is still ongoing, and will likely continue for some time.

 

 

EXAONE Lab, led by Jinsik, is in charge of creating a super-giant language model to be used for challenges faced by LG Group affiliates and partner companies. There are two challenges that need to be solved in order to utilize super-giant language models in the actual field. One is the necessity of securing a level of accuracy (performance) that can be utilized in the product, and the other is the necessity of the model being lightweight and efficient to minimize operational costs. For this, EXAONE Lab is intensively researching language model learning techniques to ensure high accuracy, and meanwhile, we are also working on making the language model lighter.

Recently, in collaboration with LG Electronics and LG Household & Health Care, we have developed an AI service that aids customer consultation. This function converts customer consultation details into text in real-time, analyzes the consultation details, and provides the necessary information to the consultant. A function that summarizes the details of the consultation and organizes them for later use after the consultation is over is included.

With LG U+, we have developed a technology that analyzes the opinions of customers about services in detail by field, and they are being used to improve services. For example, it analyzes what is the price aspect, what is the design like, etc. With Woori Bank, one of our partner companies, we are developing technology to perform banking tasks through conversations with AI bankers in digital kiosks.

 

EXAONE, a super-giant AI that simultaneously understands language and images

 

 

The name of LG’s super-giant AI model, EXAONE, stands for “Expert AI for EveryONE,” and aims to become an expert AI. In other words, the aim is to use it as a tool when a human is doing something to create a higher level of output by instructing the AI to do what AI can do best. In particular, EXAONE possesses the ability to understand both language and image simultaneously. Because it is able to think by linking language and images, it is able to take a sentence as the input and draw a corresponding image, or conversely, take an image such as a photograph as the input and explain it in a sentence.

As the size of the models increase or data is added to them, super-giant AI is showing new abilities that have not been seen before. Therefore, Jinsik predicts that it is highly likely that research on increasing the size of models and data is to be conducted for the time being. He also talked about the challenges that need to be addressed at the same time.

Q. What are the future prospects of super-giant AI research?

"As an increase in model and data size means an increase in learning or operation costs, the main point seems to be creating an optimized model that can achieve high performance while taking the size of the model and data appropriately. What's more, the data used for the learning of super-giant AI may contain biased or incorrect information, and because of this, ethical issues may arise. Therefore, it's expected that active research will be conducted on how to solve these issues."

 

The competitiveness of EXAONE Lab lies in its elite members, composed of professional scientists and engineers!

 

 

Jinsik has a special affection and pride for the members of EXAONE Lab. He proudly explains that all of the members are elite members with incredible professionalism and passion.

Q. As the leader of EXAONE Lab, please show off the members.

"Every single member comes together with passion and ambition. When they have a question, they're professional scientists who persistently dig until they find the answer they're looking for, and they're also professional engineers who pursue perfect and neat work-- this applies to all of them. In particular, they have valuable experience in handling super-giant AI models that can't be gained elsewhere, and I'd like to boast that with super-giant AI models, every single one of them is an elite member who possesses a very high level of expertise in their respective field."

As he leads EXAONE Lab, Jinsik emphasizes growth and responsibility to the members. This is because large-scale infrastructure and data are used and a lot of resources are invested to handle super-giant AI models. To share these valuable experiences among the members and help everyone grow faster, from experimental design to process and results, the lab transparently shares and conducts research and development through intense discussions.

 

The “AI technology that benefits the lives of humans” that EXAONE Lab will seek to create

 

 

Q. What is the ultimate vision for EXAONE Lab?

“EXAONE Lab’s ultimate goal is to develop AI technologies that can benefit the lives of humans. With the development of language model technology, AI is able to easily acquire knowledge expressed in language, and in the future, language models will be able to hold a lot of knowledge that’s present in the world. Based on this, I believe that AI can play a role in helping humans make optimal decisions. In order to realize this, in particular, the technology to effectively incorporate knowledge into models and to construct and infer related knowledge must be strengthened.

To join the EXAONE Lab, you’ll need a high level of understanding of super-giant language models and the underlying technologies. It'd also be nice to have a sense of purpose in developing AI technology that helps humans through the continued interest in super-giant AI.”

 

▶ Watch the YouTube Video of Jinsik’s Interview (Link)