DeepSeek
36 articles found in this topic.
Clawdbot AI Agent's Rapid Adoption Raises Security Concerns Over Hidden Malicious Code
Clawdbot, an AI agent with rapid adoption, raises security concerns due to its autonomous capabilities and potential for hidden malicious code. Despite its versatility and accessibility, a new 'Invisible Jailbreak' vulnerability highlights significant risks.
Hilight AI Platform Enables Consistent Product Video Generation for E-commerce
Hilight AI, a new platform by Yingcai AI, revolutionizes e-commerce video production by generating consistent, high-quality product videos from a single link. It addresses historical AI video generation challenges, ensuring product integrity and consistency across frames. Hilight automates scriptwriting, digital human matching, and rendering, making AI-generated videos commercially viable.
DeepSeek-OCR2 Introduces "Causal Flow" Visual Reasoning, Outperforms Gemini Pro
DeepSeek has open-sourced DeepSeek-OCR2, an advanced OCR model featuring the DeepEncoder V2 and a "Causal Flow" visual reasoning approach. This innovation allows the AI to process documents with human-like logic, dynamically adjusting reading order based on content semantics. DeepSeek-OCR2 sets new benchmarks, outperforming previous models and even closed-source competitors.
Prompt Repetition Significantly Boosts Non-Reasoning LLM Accuracy, Google Research Shows
A new Google study reveals that simply repeating a prompt can significantly enhance the accuracy of non-reasoning large language models (LLMs), with some models showing accuracy jumps from 21.33% to 97.33%. This "copy-paste" method proved effective across various benchmarks and models, outperforming single-prompt queries in 47 out of 70 tasks.
DeepSeek V4 Model Expected to Launch in February with Enhanced Programming Capabilities
DeepSeek is set to release its V4 model in mid-February, promising significant advancements in programming capabilities. Internal testers report V4 as a major leap, aiming to surpass top models like Claude and GPT. This release follows DeepSeek's successful R1 model and continued iterations, setting high expectations for its performance.
Chinese AI Leaders Discuss Model Differentiation and Future Paradigms at AI NEXT Event
Top Chinese AI figures from Zhipu AI, Kimi, Qwen, and Tencent convened at AI NEXT to discuss the future of AI. They explored model differentiation, the shift from chat paradigms to 'Action' and AI agents, and the divergence between consumer and business applications. The discussion highlighted that future AI competition will depend on models reflecting specific values and worldviews.
Epoch AI Report: Chinese AI Models Lag US Counterparts by Seven Months
A new Epoch AI report reveals Chinese AI models lag US counterparts by an average of seven months in capability. This disparity is linked to open-source versus closed-source development, with US models showing rapid advancements while Chinese models exhibit a 'leapfrog' catch-up pattern.
DeepSeek R1 Paper Expands to 86 Pages, Details Reinforcement Learning for AI Reasoning
DeepSeek has released an 86-page paper for its R1 model, detailing how AI reasoning can be enhanced through reinforcement learning. The paper provides a reproducible technical report, including data recipes, infrastructure, training costs, and performance comparisons with models like OpenAI o1 and GPT-4o.
Jensen Huang Showcases DeepSeek, Kimi AI Models for Next-Gen Chip Performance
Nvidia CEO Jensen Huang showcased Chinese AI models DeepSeek and Kimi K2 at CES 2026, demonstrating the capabilities of the upcoming Rubin architecture. These models, utilizing the Mixture of Experts approach, showed significant performance gains and cost reductions, highlighting the growing impact of Chinese AI on global benchmarks.
CAS Launches VTCBench to Evaluate Visual Language Models in Long Text Compression
CAS introduces VTCBench, a new benchmark to evaluate visual language models (VLMs) in long text compression. It assesses cognitive limits through retrieval, reasoning, and memory tasks, revealing a "U-shaped curve" in model performance and spatial attention bias in VLMs.
Ant Group Open-Sources AntAngelMed Medical AI Model, Tops Global Benchmarks
Ant Group has open-sourced its AntAngelMed Medical Large Model, a generative AI system for medical professionals and the public. Developed with Zhejiang Provincial Health Commission, it tops global medical AI benchmarks like OpenAI's HealthBench and MedAIBench, showcasing leading capabilities in medical knowledge, reasoning, and ethics.
DeepSeek Introduces Manifold-Constrained Hyper-Connections for Enhanced AI Model Stability
DeepSeek introduces Manifold-Constrained Hyper-Connections (mHC), a new method to enhance the stability and efficiency of information flow in large AI models. mHC addresses issues like signal explosion and vanishing through a double stochastic matrix constraint, significantly reducing signal distortion during model training. This innovation aims to prevent costly training failures.