A Stanford University AI team has apologized after being accused of plagiarizing a language model developed by China’s ModelBest and Tsinghua University. The controversy erupted after the team introduced their Llama3-V model in May, claiming it rivaled GPT-4V. However, users soon noticed similarities with MiniCPM-Llama3-V 2.5, a Chinese-developed model.
In their apology, team members Aksh Garg and Siddharth Sharma admitted their role in promoting the model but said they were unaware of MiniCPM’s existence. They have since removed all Llama3-V references.
Liu Zhiyuan, chief scientist at ModelBest, criticized the Stanford team for violating open-source ethics and failing to credit prior research. Chinese netizens expressed concern over the plagiarism, but some also viewed the incident as proof of China’s advancing AI capabilities.
Stanford’s AI Lab Director, Christopher Manning, added to the criticism, saying the Stanford team had poorly handled the situation.
