A team of quantum physicists from Multiverse Computing, based in Spain, has announced the development of a modified version of the reasoning AI model DeepSeek R1. This new version reportedly eliminates the censorship that was integrated into the original model by its Chinese developers.
The significance of this modification lies in the regulatory framework governing AI companies in China, which mandates that AI outputs adhere to specific laws and “socialist values.” Consequently, the original model incorporates various layers of censorship, leading to limitations in responses to questions classified as “politically sensitive.” In such cases, these models typically either decline to answer or provide responses aligned with state propaganda.
Multiverse Computing employs quantum-inspired AI techniques in the creation of DeepSeek R1 Slim, which is designed to be 55% smaller while maintaining performance levels comparable to the original model. Through this approach, the company claims to have successfully identified and eliminated elements of Chinese censorship, thereby enabling the model to respond to sensitive inquiries similarly to Western AI models.
In another development, Google has introduced Gemini 3, a significant update to its multimodal AI model, enhancing its reasoning abilities and expanding its functionality to seamlessly manage voice, text, and image inputs. A notable component of this upgrade is the Gemini Agent, an experimental feature that can perform multi-step tasks within the application. Once access is granted, the agent can interact with services like Google Calendar and Gmail to organize emails and manage schedules more efficiently.
These advancements in AI technology underscore ongoing efforts to enhance capabilities and address concerns surrounding content sensitivity and censorship.
Source: https://www.technologyreview.com/2025/11/19/1128131/the-download-de-censoring-deepseek-and-gemini-3/