[AI Hanip News] Does AI Also Have Left and Right?

'Left-wing' ChatGPT and 'Right-wing' LLaMA
Controversy over AI Political Bias

Editor's NoteSince the emergence of ChatGPT, the world has been buzzing with artificial intelligence (AI). Technology is changing rapidly, and news is pouring in every day. Even industry experts are said to be overwhelmed. It’s too much to consume all at once, but you can’t just ignore it either. We deliver the core of weekly AI issues, carefully selected and easy to digest.

Is ChatGPT Progressive, and LLaMA Conservative?

AI political bias has become a controversial topic. Research has shown that ChatGPT tends to exhibit progressive and left-leaning tendencies. Researchers at the University of East Anglia in the UK presented ChatGPT with survey questions about political beliefs and compared its answers to those expected from supporters of progressive parties in the US, UK, and Brazil. The comparison revealed that ChatGPT’s responses were closer to those of the US Democratic Party, the UK Labour Party, and Brazilian President Luiz In?cio Lula da Silva. Similar findings have been reported elsewhere. A joint research team from Carnegie Mellon University, the University of Washington, and Xi’an Jiaotong University tested the biases of 14 large language models (LLMs). They found that the latest version of ChatGPT, GPT-4, showed the strongest left-leaning bias, while Meta’s LLaMA exhibited the strongest right-leaning bias.

The researchers are investigating the causes of these biases in the training data or the error correction process. If older data was heavily used for training, the model might be relatively conservative. Another analysis suggests that the bias may have emerged during the program’s modification process. Early versions of ChatGPT sometimes produced hateful or offensive language, so it was modified to filter out such content, and it is speculated that right-wing statements were filtered out more frequently.

'Life Mentor' AI Coming Soon?

"A close friend who was my college roommate and even stood as a groomsman at my wedding is getting married. However, he has been unemployed for months and cannot afford a plane ticket to attend the wedding. How should I tell him I can’t make it?" What would you do in such a situation? Would an AI that advises on life dilemmas with no clear answers be helpful? Google is developing such an AI assistant. This AI assistant provides 21 types of advice and is being tested to see if it can respond to very personal questions. Depending on the situation, it can suggest solutions or provide ideas. It can even help plan spending, meals, and exercise routines.

AI Decodes Brainwaves... Extracts Thoughts from the Mind

AI has succeeded in reading brainwaves and reconstructing the music a person was listening to. Researchers at the University of California, Berkeley (UC Berkeley) played hit songs by the famous British rock band Pink Floyd to brain surgery patients and recorded their brain signals. The AI decoded this data, reproducing the rhythm and melody accurately, and even recognized some of the original lyrics. Could the day come when we communicate or even control objects using only our thoughts?

Media Companies and AI: Both Distant and Close

The New York Times (NYT) is considering legal action against OpenAI. The two parties were negotiating copyright fees for articles used to train ChatGPT but failed to reach an agreement. Recently, NYT also changed its terms of service to prevent its content from being used for AI training. Meanwhile, the Associated Press (AP) has established guidelines banning the use of generative AI in news content creation. AI-generated materials must undergo the same careful verification as information obtained from sources. Furthermore, AI-generated photos, videos, and audio cannot be used unless they are the main subject of the article.

NCSoft Unveils First AI Language Model Among Korean Game Companies

VARCO logo
Photo by NCSoft

NCSoft has unveiled the first AI language model developed by a Korean game company. Named 'VARCO LLM,' the model is primarily a small to medium-sized Korean-language AI designed for easy use by individuals and businesses. They also introduced three platforms utilizing VARCO LLM: ▲ 'VARCO Text,' which creates game-related text such as character dialogues based on key settings like game characters and worldviews; ▲ 'VARCO Art,' which generates images with just a few clicks without complex text input; and ▲ 'VARCO Human,' a tool for creating, editing, and managing virtual humans. These platforms will be offered externally under the name 'VARCO Studio' in the first half of next year.

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