Scientists at Multiverse Computing, a Spanish firm specialising in quantum-inspired AI techniques, created DeepSeek R1 Slim, a model that's 55% smaller than the original one—and is also censorship-free. Multiverse used a complex approach borrowed from quantum physics that uses networks of high-dimensional grids to represent and manipulate large data sets, which allowed them to identify and remove specific bits of information with precision.
DeepSeek R1 Slim has 300 billion fewer parameters than the original model, directly halving memory consumption and deployment costs whilst still maintaining accuracy on all deep reasoning tasks. Multiverse's compression technique, CompactifAI, is based on quantum-inspired tensor networks, which allows them to identify and remove parameters that contribute little to the model's overall performance. It additionally allows them to isolate and remove weights tied to specific learned behaviours, such as censorship, without degrading the model's core knowledge. To test the results, the researchers compiled a data set of around 25 questions on topics known to be restricted in Chinese models, and the uncensored model was able to provide factual responses comparable to those from Western models.
In January 2025, DeepSeek R1 gained global recognition for its exceptional performance across mathematics, coding, and logical reasoning benchmarks, rivalling or surpassing its contemporaries, and it quickly gained traction for both its raw power and open-source nature. Multiverse's breakthrough demonstrates that that quantum-inspired techniques can selectively eliminate political bias and make models even more accessible whilst preserving core analytical capabilities.
Sources:
https://multiversecomputing.com/resources/deepseek-r1-uncensored-full-power-fraction-of-the-size
https://www-technologyreview-com.cdn.ampproject.org/c/s/www.technologyreview.com/2025/11/19/1128119/quantum-physicists-compress-and-deconsor-deepseekr1/amp/
