Kakao Brain Publishes Poetry Collection Using Ultra-Large AI Language Model ‘KoGPT’

[Asia Economy Reporter Seungjin Lee] Kakao Brain announced on the 1st that it has developed a poetry-writing artificial intelligence (AI) model called ‘SIA’ (hereinafter Sia) in collaboration with the media art group Slitscope, and will publish its first poetry collection titled ‘The Reason for Writing Poems.’

‘Sia’ is an AI model that writes poetry based on Kakao Brain’s large-scale AI language model KoGPT. It learned poetic techniques by reading over 13,000 poems. When keywords and commands are input, ‘Sia’ understands the context of the given information and immediately composes a poem.

The poetry collection ‘The Reason for Writing Poems,’ which will be published on the 8th, consists of a total of 53 poems. Using the machine language 0 and 1 for digital computation, which is Sia’s language, the theme of Part 1 is “Gong (0),” and the theme of Part 2 is “Il (1).” The use of ‘Gong’ instead of ‘Zero’ reflects the intention of Kakao Brain and Slitscope to encompass the relationship between existence and non-existence, meaning and meaninglessness.

‘Gong (0),’ which carries the meaning of meaningless and non-existence (off), includes poems generated from arbitrary expressions presented as poetic themes derived from Slitscope’s previous work notes. ‘Il (1),’ which carries the meaning of meaning and existence (on), includes poems themed around mathematics and science.

Slitscope is a media art group founded in 2018 by director and media artist Jemin Kim and AI researcher and software developer Geunhyung Kim, who have worked in the distinct fields of art and science. They began with the performance ‘I Question,’ where AI and artists improvised text creation together.

In this project, Slitscope collected over 13,000 poems and further trained Kakao Brain’s KoGPT model to develop ‘Sia.’ They also generated poems in various tenses, refined them, and ultimately selected 53 poems for the collection.

Slitscope plans to hold a poetry play performance using the poems included in ‘The Reason for Writing Poems’ at the Daehangno Arts Theater Small Theater from the 12th to the 14th.

Meanwhile, KoGPT, the Korean language-specialized large-scale AI language model developed by Kakao Brain, was built based on 6 billion parameters and 200 billion tokens of Korean data, enabling it to understand Korean both lexically and contextually. Last November, the KoGPT model was released on the open-source community GitHub.

Kakao Brain CEO Ildoo Kim said, “Through the publication of this poetry collection, we were able to confirm the infinite artistic possibilities of KoGPT,” and added, “We plan to continuously explore various cultural and artistic fields where Kakao Brain’s large-scale AI models can be applied.”

This content was produced with the assistance of AI translation services.