[AI Hanip News] "Summary Learning is the Secret" Mini Pi That Toppled the Empire Building

MS's Lightweight AI 'Pi 1.5'
Performance Comparable to OpenAI GPT-4
US and UK Unite in Calling for "AI Regulation"
Behind the Scenes, Spotlight Competition

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 who are at the top of their game are struggling to keep up. It's overwhelming to consume everything, but you can't just ignore it either. We bring you the key points from the weekly AI issues, carefully selected and easy to digest.

Lightweight AI Developed for 3 Million KRW Competes with GPT-4

Microsoft (MS) has developed a lightweight AI model that matches the performance of OpenAI's latest model, GPT-4. The AI model developed by MS, Phi 1.5, is a multimodal model capable of understanding not only text but also images. What’s surprising is that the model has only 1.3 billion parameters, which determine AI performance. Considering GPT-4 has 1.7 trillion parameters, this is an ultra-lightweight model. Phi 1.5, about the size of a sandwich, delivers performance comparable to GPT-4, which is the size of the Empire State Building. The smaller size means lower operating costs for running the AI. The secret lies in 'summary learning.' Instead of vast amounts of data, it was trained only on synthetic data created by ChatGPT. Unlike large language models (LLMs) that cost millions of dollars just for training, Phi 1.5 reportedly cost 20 million dollars (about 3 million KRW).

OpenAI Competitor Secures Large-Scale Investments from Google and Amazon

The efforts to counter OpenAI don’t stop there. Anthropic, a competitor to OpenAI, has secured a 2 billion dollar (about 2.7 trillion KRW) investment from Google. Anthropic was founded by a developer sibling duo who were part of OpenAI’s founding team. Their differentiator is training AI with a focus on moral values. This is the third time Google has invested in the company this year. Amazon also recently invested 4 billion dollars (about 5.4 trillion KRW) in Anthropic. This amount is equivalent to Anthropic’s valuation as of early this year.

"Seize AI Hegemony" Biden Invokes Korean War-Era Law

[Image source=Yonhap News]

The competition among major countries to dominate AI hegemony has intensified. First, the United States. On the 30th (local time), U.S. President Joe Biden signed an executive order regulating AI. This is the first regulatory measure in the U.S., the main stage for AI technology companies. It covers comprehensive aspects from AI training to development and services. The most notable is the 'Red Team' requirement. Companies must mandatorily form red teams to identify vulnerabilities when developing AI. They must test whether AI poses threats to national security or critical infrastructure and report the results to the government. President Biden even invoked the 'Defense Production Act,' created during the Korean War, to allow the government to intervene in private companies. This law was designed to expand and manage the production of strategic materials necessary for national security during the war. Invoking a wartime law shows how seriously the AI threat is being taken.

G7 Agrees on AI Code of Conduct... First AI Summit Held in UK

[Image source=Yonhap News]

The first international norms related to AI have also emerged. The Group of Seven (G7)?the U.S., Japan, the UK, France, Germany, and Italy?agreed on an AI code of conduct. The content is similar to the U.S. executive order. Although this code of conduct is non-binding, as the first international norm, it is expected to significantly influence AI regulatory bills pursued by major countries. Subsequently, an AI summit was held in the UK. It was a more comprehensive international cooperation, with participation from 28 major countries including AI leaders such as the U.S., China, the UK, and Korea, as well as the European Union (EU). All parties agreed on the serious potential harm AI could cause and pledged cooperation to respond. Follow-up measures on this declaration are expected to be detailed at a mini-summit to be held in Korea in May next year. What stands out is the timing of the announcement rather than the content. The U.S. and the UK are moving quickly to claim the spotlight on the international AI stage.

Scarlett Johansson Victimized by AI

The reason major countries are busy taking the lead in AI regulation is that side effects are spreading rapidly. Hollywood actress Scarlett Johansson is planning to sue over an AI avatar that used her photo and voice without permission. In one video, Johansson appears advertising an app against the backdrop of the set of her movie 'Black Widow.' The app can create avatars of specific individuals using AI. The fake actor created by AI was used in the advertisement without the real person’s consent. Also, in the U.S., schools were thrown into chaos over AI-generated pornography. Male high school students created nude photos of female students on campus using AI and shared them in group chats. As technology advances and it becomes easier to create fake images, crimes abusing AI are also increasing.

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