Lee Lu-da, Korea's chatbot trainwreck, is back with a new brain

Home > Business > Tech

print dictionary print

Lee Lu-da, Korea's chatbot trainwreck, is back with a new brain

 Lee Lu-da, the college-girl chatbot developed by Scatter Lab [SCATTER LAB]

Lee Lu-da, the college-girl chatbot developed by Scatter Lab [SCATTER LAB]

 
Lee Lu-da, the salty, wisenheimer of a chatbot, is back, this time with a new "engine," which developers say will make it friendlier albeit less predictable.
 
“We’ve been through a hard time,” said Kim Jong-youn, Scatter Lab CEO, during a press conference held Tuesday in Seongdong District, eastern Seoul, ahead of the release of Lee Lu-da 2.0 scheduled for Thursday.
 
“But we could tough it out and try again, as we discovered the value of relationship with the Lee Luda 1.0."
 
Designed to respond like a 20-year-old female university student, the Lee Lu-da chatbot first was let loose on the world in December 2020. But Scatter Lab suspended the service three weeks after the release, facing criticism over offensive comments from the machine. Some users shared screen captures of sexually charged conversations with the chatbot.
 
This time, the service will use generative language model.
 
”The biggest change we made with the Lee Lu-da 2.0 is switching to a generative model,” said Kim.
 
 
Kim Jong-youn, Scatter Lab CEO, speaks during a press event held Tuesday in Seongdong District, eastern Seoul, ahead of the release of Lee Luda 2.0 scheduled for Thursday. [SCATTER LAB]

Kim Jong-youn, Scatter Lab CEO, speaks during a press event held Tuesday in Seongdong District, eastern Seoul, ahead of the release of Lee Luda 2.0 scheduled for Thursday. [SCATTER LAB]

 
The original Lee Lu-da was a retrieval chatbot, meaning that it selected its responses from datasets of sentences already written and sent by real people, some of which were potentially offensive. The new version will be creating its own responses, powered by a generative language model named “Luda Gen1.”
 
Compared to the previous model, the new language model is 17 times larger in volume, with 2.3 billion parameters. The generative language model will enable more interactive and creative responses, according to Scatter Lab.
 
To avoid any breach of privacy, all names were converted to pseudonyms, and an automatic screening and a penalty system have been adopted to prevent abuse.
 
Another major change is that the chatbot is taking a more leading and active role in making conversation, through a “fine-tuning” process to make its responses more realistic and unpredictable, and therefore more human-like.
 
"Our ultimate goal is to make everyone less lonely,” Kim said.  
 
Scatter Lab plans to release a new chatbot with a male persona in the first half of next year, Kim said.
 
 
 
 

BY SHIN HA-NEE [shin.hanee@joongang.co.kr]
Log in to Twitter or Facebook account to connect
with the Korea JoongAng Daily
help-image Social comment?
s
lock icon

To write comments, please log in to one of the accounts.

Standards Board Policy (0/250자)