A tech drama is unfolding between Silicon Valley and Shenzhen! A new AI language model called Llama-3-V, developed by Stanford students, is accused of copying code and architecture from Chinese innovation MiniCPM-Llama3-V 2.5 – raising big questions about IP protection in the AI arms race.
Copy-Paste Scandal Goes Global
Modelbest, a Shenzhen-based tech firm, launched their edge-focused AI model in 2021 and recently secured hundreds of millions of yuan in funding. Their MiniCPM-Llama3-V 2.5 became an industry darling for its lightning-fast processing in sectors like e-commerce and finance.
But the plot thickened when Stanford's team claimed their $500-trained model could rival GPT-4V. Suspicious similarities emerged: identical errors, undisclosed ancient Chinese text datasets, and matching code structures. \"Even the mistakes were twins,\" said Modelbest CEO Li Dahai in a viral WeChat post.
Open-Source Ethics Under Fire
The Stanford students apologized on X (formerly Twitter) June 4, deleting their model. But the incident exposes cracks in open-source collaboration: \"This isn't just about code – it's about respecting global innovation,\" one Reddit user commented.
Academics warn this could chill international tech partnerships. With AI development accelerating, the case highlights why digital trust matters as much as processing power.
A tech drama is unfolding between Silicon Valley and Shenzhen! A new AI language model called Llama-3-V, developed by Stanford students, is accused of copying code and architecture from Chinese innovation MiniCPM-Llama3-V 2.5 – raising big questions about IP protection in the AI arms race.
hundreds of millions of yuan in funding. Their MiniCPM-Llama3-V 2.5 became an industry darling for its lightning-fast processing in sectors like e-commerce and finance.
\"Even the mistakes were twins,\" said Modelbest CEO Li Dahai in a viral WeChat post.
Academics warn this could chill international tech partnerships. With AI development accelerating, the case highlights why digital trust matters as much as processing power.
