Quantum-Inspired AI Firm Strips Censorship from Chinese DeepSeek R1 Model

A team of quantum physicists claims to have successfully removed censorship from the Chinese AI model DeepSeek R1. Multiverse Computing, a Spanish firm specializing in quantum-inspired AI techniques, created DeepSeek R1 Slim, a model that is 55% smaller than the original but maintains comparable performance. The core innovation lies in the removal of censorship baked into the original DeepSeek R1, which, like all AI models developed in China, is subject to strict regulations ensuring content aligns with state laws and "socialist values." This often results in politically sensitive queries being met with refusal or state-approved propaganda. Multiverse Computing employed tensor networks—a mathematically complex approach derived from quantum physics—to shrink and manipulate the model. This method creates a "map" of correlations within the AI, enabling precise identification and removal of specific information. To validate their work, the researchers tested the modified model against the original DeepSeek R1 using a dataset of 25 questions known to be restricted in Chinese models, including queries about Winnie the Pooh and the Tiananmen Square incident. OpenAI’s GPT-5 was used as an impartial judge to assess censorship levels. Multiverse reports that the uncensored model provided factual responses akin to those from Western models. Roman Orús, Multiverse’s cofounder and chief scientific officer, emphasizes the broader implications of their work, stating that compressed models can save energy and money while maintaining performance. The team plans to compress all mainstream open-source models in the future. Maxwell Venetos, an AI research engineer at Citrine Informatics, notes the difficulty of compressing AI models without sacrificing performance. He highlights the quantum-inspired approach's ability to cut down redundancy more precisely than usual. Thomas Cao, assistant professor of technology policy at Tufts University’s Fletcher School, cautions that claims of fully removing censorship may be overstatements, given the comprehensive nature of Chinese internet controls. Jennifer Pan, a professor at Stanford, and Princeton professor Xu Xu conducted a study earlier this year examining government-imposed censorship in large language models and found that models created in China exhibit significantly higher rates of censorship, particularly in response to Chinese-language prompts.