A recent experiment compared the performance of two AI models, Qwen 3.8 MAX and DeepSeek V4, using the PlyLoom data visualization tool. The study aimed to evaluate how each model generates graph-based responses when tasked with describing its competitor.
The Qwen 3.8 MAX model provided a summary of its competitor's release history and key technologies, specifically highlighting training costs. However, the resulting Mermaid diagram was limited in scope, and the model created duplicate entities across different data sheets.
DeepSeek V4 adopted a different approach, focusing on questions and potential risks rather than definitive statements. The model explained that it lacked specific source materials regarding Qwen, as the input field for documentation was left empty in the test.
DeepSeek V4 noted that it adheres to a contract requiring it to avoid inventing facts or dates. It proposed three operational modes for PlyLoom: a strict mode based only on provided sources, a draft mode using general knowledge with unverified labels, and a mixed mode.
The experiment concluded with a parity between the two models, as each handled the lack of provided source data differently. Users can test their own data and theories using the PlyLoom platform at its official website.
