Tech Trends 📰

DeepSeek faces ban from Apple, Google app stores in Germany

Germany has asked Apple and Google to remove the DeepSeek AI app from their app stores, claiming it illegally sends users’ personal data to China.. The German data protection commissioner said DeepSeek failed to prove that user data stored in China is protected to EU standards.. DeepSeek has already faced bans in Italy and on government devices in the Netherlands, with Belgium and Spain also raising concerns.. The app’s privacy policy confirms it stores requests and uploaded files on servers in China, which raises national security worries.. U.S. lawmakers are also planning to ban the use of Chinese-developed AI models like DeepSeek in government agencies, especially after reports of its links to China’s military and intelligence..

DeepSeek may have used Google's Gemini to train its latest model

DeepSeek’s new R1-0528 model shows signs it may have been trained on Google Gemini outputs, according to AI developers.. The model uses language and reasoning patterns similar to Gemini, raising suspicions of unauthorized training data use.. DeepSeek has previously been linked to distillation techniques involving OpenAI’s models, including ChatGPT.. Companies like OpenAI, Google, and Anthropic are strengthening security to prevent such model copying.. It’s becoming harder to filter out AI-generated content from training datasets due to widespread “AI slop” online..

DeepSeek updates its R1 reasoning AI model, releases it on Hugging Face | TechCrunch

DeepSeek has released an updated version of its R1 reasoning AI model on Hugging Face.. The update is described as a “minor” improvement and is available under the permissive MIT license.. The model is huge, with 685 billion parameters, making it unsuitable for regular consumer hardware.. The Hugging Face repository includes only configuration files and weights—no detailed description.. DeepSeek’s original R1 model gained attention for rivaling OpenAI, and it has raised concerns among U.S. regulators over national security..

DeepSeek upgrades its math-focused AI model Prover | TechCrunch

DeepSeek released Prover V2, an upgraded AI model focused on solving math proofs and theorems.. It is built on the V3 model, which uses a 671 billion parameter Mixture of Experts (MoE) architecture.. A distilled version of Prover V2 was also uploaded on Hugging Face for developers.. DeepSeek is expanding and may soon update its R1 reasoning model after upgrading its general-purpose V3 model..