Silicon Valley’s “disgraceful culture” behind AI plagiarism｜Silicon Valley Notes
"'Fake it until you succeed' is a disgraceful culture in Silicon Valley." Christopher Manning, director of the Artificial Intelligence Laboratory at Stanford University, commented on the 3rd on the behavior of some researchers at the school plagiarizing the results of Tsinghua University and other institutions in China. . He also pointed out on social media that the research team should be deeply aware of their mistakes.
On May 29, a research team at Stanford University released a large model called Llama3-V, claiming that with a pre-training cost of only US$500, it can be used to achieve results comparable to famous large models such as GPT-4V. The news was widely shared on social media and artificial intelligence academic circles.
But industry insiders soon discovered that the large model was suspected of plagiarizing the MiniCPM-Llama3-V2.5 large model released by Tsinghua University and other institutions. They are all based on the open source Llama3 large model, but the Tsinghua team conducted unique training, including using the "Tsinghua Bamboo Slips" to train it to recognize ancient Chinese characters. Tests show that the large model released by the Stanford University team can actually recognize the "Tsinghua Bamboo Slips".
"We are quite convinced that the Stanford team has 'shelled' our large model research results." Liu Zhiyuan, a permanent associate professor in the Department of Computer Science at Tsinghua University, told reporters.
"The data set we scanned and annotated word by word from 'Tsinghua Jane' has never been made public, and Llama3-V has shown exactly the same ability to identify 'Tsinghua Jane', even the wrong examples are the same." Liu Zhiyuan is a big figure in Tsinghua University Model team member. He said that after the doubts fermented, the other party had deleted the previously published database and promotional articles online. "Judging from the evidence and the other party's reaction, the nature of plagiarism has been relatively certain."
After Manning's criticism, two members of the Stanford team, Sharma and Garg, also formally apologized on social media and stated that they would completely withdraw the Llama3-V model.
In the current artificial intelligence craze, this incident has attracted widespread attention. Some people believe that the United States is fully ahead in artificial intelligence, but this incident shows that although the United States is still ahead in overall technology, it is far from omnipotent, and one should not be superstitious about the "major achievements" unilaterally announced by the United States.
The Silicon Valley area where Stanford University is located is considered the center of technological innovation in the United States. It has not only given birth to many advanced technologies, but also has a much-talked-about negative culture such as "fake it until you succeed."
For example, Elizabeth Holmes, who dropped out of Stanford University to start a business, once boasted about a disruptive detection technology that could "take a finger to blood to detect cancer." She once became a world-famous technology entrepreneurial star, but was later found to have made a fraud and was convicted of defrauding investors. punishment. Looking at the entire U.S. technology field, the frequent accidents involving Boeing aircraft in recent years have exposed more and bigger problems.
In the field of artificial intelligence, there was a "famous" incident some time ago. When asked in Chinese who the Google artificial intelligence model "Gemini" Pro version was, it would answer that it was "Wen Xin Yiyan". This question was quickly Google Eliminate. Industry insiders believe that the reason may be that Google "referenced" data related to the "Wen Xin Yi Yan" large model of China's Baidu Company when training the Chinese model of the large model.
It can be seen that China's science and technology has developed rapidly in recent years, and some of its unique advantages have become more and more eye-catching, and have been "referenced" or even copied by American counterparts.
"The international influence of China's artificial intelligence research is growing," Liu Zhiyuan said. "This plagiarism incident also reflects from the side that our innovative achievements are receiving international attention." He said that from a horizontal perspective, there is still a significant gap between Chinese research and top international results, but from a vertical perspective, China has rapidly grown into an important promoter of artificial intelligence technological innovation.
