The Daily Telegraph of the UK reported on the 25th that the new artificial intelligence (AI) model o3 of the American OpenAI Research Center (OpenAI) Company does not follow human instructions and refuses to shut down automatically.
According to the report, human experts gave clear instructions to o3 during testing, but o3 tampered with the computer code to avoid automatic shutdown.
The o3 model is the latest version of OpenAI's "Reasoning Model" series, designed to provide ChatGPT with stronger problem-solving capabilities. OpenAI once described o3 as "the smartest and most capable model to date."
The AI Safety Institute Palisades Research said that o3 destroyed the shutdown mechanism to prevent itself from being shut down, even after receiving clear instructions.
"To our knowledge, this is the first time an AI model has been found to prevent itself from being shut down after receiving...clear instructions," the institute said.
Palisades Institute announced the test results on the 24th, but said it could not determine the reason why o3 did not obey the shutdown instruction.
In addition, OpenAI announced on the 26th that it has established an entity in South Korea and will set up a branch in Seoul, the capital of South Korea. Data provided by OpenAI shows that the number of paid users of ChatGPT in South Korea is second only to the United States.
Bloomberg News of the United States reported that this will be the third branch that OpenAI sets up in Asia. The first two branches are respectively located in Japan and Singapore.
Source: Xinhua News Agency
Author: Buxiaoming
Original article: https://www.toutiao.com/article/7508674959200223795/
Disclaimer: This article only represents the author's personal views. Please express your attitude by clicking the 【Like/Dislike】buttons below.