Two Stanford University computer science undergraduates have apologised on X following allegations that their recently made AI model, Llama 3-v, resembled the structure and code of a previously crafted model by Chinese scientists. The AI model faced intense scrutiny following the successful launch of Llama 3-v last week. This drew comparisons to MiniCPM-Llama3-V 2.5, which is a model crafted and developed by Tsinghua University’s Natural Language processing lab and the AI startup ModelBest, based in Beijing.
Llama 3-V gained widespread attention for its outstanding performance at the launch. However, this raised concerns and an enquiry into its originality. This led to developers acknowledging its striking resemblance to the previously developed Chinese AI model. The Stanford computer science graduates, Aksh Garg and Shiddarth Sharma publicly apologised on X to the authors of MiniCPM and wrote, “We sincerely apologise to the authors of MiniCPM for our failure to verify the originality of Llama3-V.”
Furthermore, the students stated, “Mustafa, who wrote the code, described exciting extensions that we promoted without knowing about the prior work by OpenBMB (founded by Tsinghua University and ModelBest Inc). We take full responsibility for this oversight. We’ve removed all references to Llama3-V in respect to the original authors.” On June 2nd, A Chinese company, ModelBest Inc. confirmed that the Stanford project Llama3-V, which displayed similar structure and code to the Chinese model, reflected the capability to identify ancient characters from China’s Warring States Period, particularly those of Qinghua Jian.
The character data observed in Llama3-V revealed identical errors to the original data. This pointed out a clear act of plagiarism as this data used by Stanford University was not publicly disclosed and was obtained over months of scanning and manual annotation of each Qinghua Jian character. The director of Stanford Laboratory, Christopher David Manning, released a statement condemning the act of plagiarism and commended the Chinese open-source model MiniCPM.
ModelBest chef Inc’s scientist, Liu Zhiyuan, who is also a tenured associate professor at Tsinghua University, told a Yicai reporter that the advancement of artificial intelligence is dependent on the worldwide exchange of data, algorithms and models. He said, “Two out of the three members of this Llama3-V team are merely undergraduate students at Stanford University, and they have a long journey ahead. If they can acknowledge their mistakes and make amends, it would be a great virtue.”
Additionally, ModelBest Inc CEO Li Dahai said, “While it’s good to be recognised by international teams, we believe in building a community that’s open, cooperative and trustworthy. We want our team’s work to be noticed and respected but not in this manner.” ModelBest Inc was established in August 2022 and in April secured, a new round of financing totalling hundreds of millions of RMB.
The investment was made by Huawei’s Hubble Technology Venture Capital with further contributions from Beijing Artificial Intelligence Industry Investment Fund, Chunhua Capital and the Chinese platform Zhihu. In February, ModelBest Inc. introduced the open-source model MiniCPM to the public.
ALSO READ: OpenAI’s Next Frontier Model: A Path to Artificial General Intelligence
