SEOUL: LG AI Research, the artificial intelligence lab under South Korea's LG Group, on Monday unveiled an upgraded version of its giant, multimodal AI model Exaone, offering enhanced performance and cost efficiency for its users.
To support the AI ecosystem, all three models of the latest Exaone 3.5 will be released as open source, enabling their use for various research purposes, according to LG AI Research.
This release comes just four months after the only lightweight model of its predecessor, Exaone 3.0, was made available as open source, reports Yonhap news agency.
To minimise the risk of generating plausible but incorrect answers and increase the accuracy and reliability of answers, the Exaone 3.5 features advanced technologies that generate answers using real-time web search results or uploaded documents.
In a technical report evaluation, the model's performance received top scores for practical usability, long-text processing, coding and math compared with global open-source AI models from the United States, China and other countries.
"As AI technology has become a major strategic asset for each country, developing AI models with our own technology is meaningful for enhancing national AI competitiveness, " an LG AI Research official said.
Last year, LG AI Research had unveiled the second generation of its giant, multimodal AI model, accelerating its efforts to become a force to be reckoned with in the increasingly competitive global AI race.
EXAONE, short for expert AI for everyone, is a giant, multimodal language model that can understand both language and images simultaneously, thus being capable of image-to-text generation and vice versa. Multimodal means having multiple modes of communicating a message.
Its deep learning algorithm can understand, recognise and predict text and learn from huge volumes of data to provide viable solutions and help answer complex questions in various fields.
LG Group established LG AI Research in 2020. Last year, the group said it would invest 3.6 trillion won ($2.8 billion) in AI research and development over the next five years.