Daily discover the most amazing AI world - from breakthrough news to innovative products, from cutting-edge projects to tech trends
A groundbreaking study on DeepSeek-R1, led by Professor Liang Wenfeng's team, has been featured on the cover of Nature. The research demonstrates how reinforcement learning enhances reasoning in large language models (LLMs), marking a significant milestone in AI development. The model's peer-reviewed validation and transparent methodology set new standards for AI research.
Kling AI has launched a groundbreaking digital human feature that transforms single images into 1-minute HD videos. The tool supports multilingual input and offers realistic lip-syncing and emotional expressions, targeting content creators, educators, and marketers.
Tongyi DeepResearch has unveiled a fully open-source AI model that outperforms many international counterparts in benchmark tests. The model, designed for deep research tasks, offers advanced reasoning capabilities and innovative data strategies while being freely accessible to the global research community.
ElevenLabs has launched Studio 3.0, a comprehensive AI-driven platform for audio and video production. The tool integrates voice synthesis, automatic music generation, sound effects, and editing features to streamline content creation for podcasts, videos, and courses. Key innovations include text-based audio editing and intelligent scene-matching background music.
Meta has launched MobileLLM-R1, a compact inference model series optimized for math, coding, and scientific reasoning. This release highlights growing enterprise preference for smaller AI models that offer better cost control, privacy protection, and specialized performance compared to large language models.
The newly released Ring-mini-2.0, a 16B parameter MoE model, activates only 1.4B parameters for inference, matching the performance of dense models under 10B. It excels in logical reasoning, programming, and math tasks, supports 128K context length, and achieves speeds over 300 tokens/s. The model is now open-source, offering efficiency and cost reduction for AI applications.
The Chinese Internet Basic Corpus 3.0, a 120GB dataset, was launched during the 2025 National Cybersecurity Awareness Week. Developed collaboratively under government guidance, it enhances AI training with improved data quality and broader sources. Available via the China Cybersecurity Association, it aims to support AI innovation.
1Password has partnered with Perplexity to launch Comet, an AI-native browser featuring advanced credential management and security features. The collaboration aims to enhance user trust and safety in AI-driven browsing experiences with end-to-end encryption and seamless auto-fill capabilities.
Elon Musk revealed major upgrades for Grok Code, including a groundbreaking 1M context window expansion. The AI tool's enhanced capabilities promise improved conversation processing and problem-solving efficiency. Additional stability fixes aim to boost reliability for developers.
Macroscope, founded by Periscope's creators, has secured $30M in Series A funding for its AI-powered code analysis platform. The tool helps developers and product leaders streamline workflows by summarizing code changes, detecting vulnerabilities, and answering technical queries in natural language.
Google's Gemini 2.5 Deep Think AI made history by securing a gold medal at the 2025 ICPC World Finals, solving a problem no human team could crack. The achievement highlights AI's growing capability in complex problem-solving and algorithmic reasoning.
Tencent's Hunyuan team has unveiled SRPO, a breakthrough technology addressing the 'oily skin' effect in AI-generated images. Developed with top universities, this method improves realism through semantic preference optimization and direct-alignment strategies, achieving significant quality improvements in just 10 minutes of training.
DeepSeek, a leading AI company, has issued a warning about fraudulent activities where scammers impersonate its officials. These criminals forge documents and solicit payments for fake services like 'computing power leasing' or 'equity financing.' The company advises users to verify information through official channels and report suspicious activities.
Meta unveiled its latest Ray-Ban smart glasses at Connect2025, featuring a built-in display and AI integration. Priced at $799, the glasses work alongside a neural wristband for advanced control. This marks Meta's push into wearable tech amid growing competition.
Tencent AI Lab has introduced Parallel-R1, a reinforcement learning framework enabling large AI models to explore multiple reasoning paths simultaneously. The innovation improves accuracy in mathematical tasks by up to 8.4% and demonstrates a 42.9% performance leap in AIME25 tests.
AI security startup Irregular has raised $80 million in a funding round led by Sequoia Capital and Redpoint Ventures, valuing the company at $450 million. The firm specializes in identifying risks in cutting-edge AI models through advanced simulation environments, positioning itself as a critical player in AI safety.
San Francisco-based Invisible Technologies has raised $100 million in a funding round led by Vanara Capital, bringing its total funding to $144 million and valuation to $2 billion. The company specializes in AI training data solutions, helping businesses deploy AI models effectively. The funds will expand its software platform and partnerships with major tech firms like AWS and Microsoft.
AI security startup Irregular raised $80M in funding led by Sequoia Capital, reaching a $450M valuation. The company specializes in predictive risk assessment for AI models through advanced simulation environments, addressing growing security concerns in the AI industry.
DeepSeek's research on its R1 large language model has been published in Nature, marking the first time an AI model passed rigorous peer review. The breakthrough highlights advancements in autonomous reasoning and sets new standards for transparency in AI research.
A new report by OpenAI reveals that 70% of ChatGPT conversations are unrelated to work, with over 700 million weekly active users globally. The study highlights growing adoption among women, seniors, and low-income users, emphasizing AI's role in daily life for education, writing, and information retrieval.
Get the latest AI news, products, and projects delivered to your inbox