DeepSeek has launched an updated version of its R1 model, the DeepSeek-R1-0528, featuring 685 billion parameters compared to 671 billion in its previous version. Unlike the proprietary models of OpenAI and Google, DeepSeek’s model weights are available to the public, promoting accessibility. The R1-0528 demonstrates enhanced reasoning capabilities, nearing the performance levels of OpenAI’s o3 and Google’s Gemini 2.5 Pro. Additionally, DeepSeek has introduced a distilled variant using Alibaba’s Qwen3 8B model, which, despite being less powerful, can effectively run on a single GPU and has outperformed other lightweight models in certain tests.
The updated model has attracted attention amid a quieter week in AI news, overshadowed by a series of events from major players in the field. Other notable announcements include Anthropic’s new voice mode for its Claude AI, allowing integration with apps like Google Workspace, and Opera’s introduction of the Neon browser, which aims to innovate the online experience with AI capabilities.
Meanwhile, concerns about job automation have arisen, with predictions suggesting AI may reduce entry-level positions significantly. Additionally, a poll indicates that many oppose a proposed ban on states regulating AI. In cultural developments, viral content and AI-driven projects, like a new humanoid robot and conservation initiatives for endangered bird species, highlight the multifaceted influence of AI technology in contemporary society.
Source link
