South Korea Bans New Users from Downloading DeepSeek

2025-02-17

South Korea's data protection authority has suspended new user downloads of the Chinese AI app DeepSeek due to non-compliance with local personal data protection laws. Services may resume once improvements are made. Italy also requested DeepSeek block its chatbot over privacy concerns. China stresses its commitment to data security and legality.

Read more

Torque Clustering: A New AI Algorithm Revolutionizing Unsupervised Learning Paradigms

2025-02-17

Researchers developed Torque Clustering, an AI algorithm that enhances autonomous learning and pattern recognition without human input. Inspired by physics' torque concept, it outperforms traditional methods with 97.7% AMI accuracy across 1,000 datasets. Fully autonomous, parameter-free, and efficient, it adapts to diverse data types and holds potential for advancing robotics, autonomous systems, and unsupervised learning in various fields. The open-source code is available for research use.

Read more

Scientists Use AI to Decode Animal Emotions

2025-02-17

Researchers are developing AI systems to analyze animal emotions and health through facial expressions. A system called Intellipig detects pain or distress in pigs, aiding farmers. Studies show AI can identify pain in horses with 88% accuracy by analyzing facial changes. These advancements use deep learning and image processing to improve animal welfare.

Read more

OpenAI Adjusts AI Model Training Strategy for a More Neutral ChatGPT

2025-02-17

OpenAI adjusts AI model training to support "freedom of knowledge," enabling ChatGPT to provide broader perspectives and reduce avoided topics. A new principle aims to avoid deception and maintain neutrality on controversial issues by presenting multiple viewpoints. While the update seeks to offer more user control, it still restricts clearly false or objectionable content. This shift may address prior criticisms but remains controversial.

Read more

Perplexity Launches Free AI Research Tool Deep Research

2025-02-17

Perplexity launched its free AI research tool, Deep Research, challenging OpenAI and Google. It generates detailed reports in minutes, with non-subscribers allowed five daily queries and Pro users getting 500. Unlike OpenAI's paid-only version, Perplexity's tool is free but limits queries for non-paying users. It scored 21.1% on the Humanity’s Last Exam and 93.9% on SimpleQA. Backed by Jeff Bezos and NVIDIA, valued at $9 billion, Perplexity faces legal issues over copyrighted content but has par

Read more

Meta Plans to Develop Humanoid Robot Hardware and Software

2025-02-17

Meta plans to develop humanoid robot hardware and software, focusing initially on household chores. Led by Mark Whitten, the project aims to create foundational AI and tech for multiple manufacturers, rather than launching branded robots immediately. Meta is in talks with Unitree Robotics and Figure AI, aligning with its Reality Labs investments. This move comes as tech giants like Tesla, Apple, and NVIDIA expand into robotics. Additionally, Meta hires John Collier for retail operations and inve

Read more

Apple Siri Upgrade Faces Delays, Possibly Postponed to iOS 18.5 Release

2025-02-17

Apple's major Siri upgrade, planned for iOS 18.4, faces delays or limited functionality due to engineering challenges and bugs. Key improvements include better app control, context-aware services, and enhanced AI capabilities. The update aims to boost Apple's competitiveness in AI but has encountered setbacks, with potential postponement to iOS 18.5. Insiders suggest some features may be disabled initially and enabled later. Apple's AI progress lags behind competitors like ChatGPT and Google Gem

Read more

Amazon Alexa AI Upgrade Delayed Due to Issues

2025-02-17

Amazon delays the launch of an upgraded Alexa by at least a month due to accuracy issues in test queries. The new version, potentially tied to a subscription service, may release on March 31 or later. Initially planned as a free trial, internal documents reveal shifting plans. Despite prior claims of completion by late 2024, challenges with generative AI capabilities persist. Meanwhile, Apple faces similar hurdles with Siri upgrades, while Google advances with its Gemini-powered assistant.

Read more

WeChat Rolls Out DeepSeek Large Model Integration in Gray-scale Testing

2025-02-17

WeChat is testing a phased rollout of the DeepSeek-R1 large language model, accessible to select users via the new "AI Search" feature. This function offers two modes: "Quick Answers" for fast replies and "Deep Thinking" for more comprehensive responses. The AI Search relies on public content and does not use personal data. Key features include authentic response styles, consecutive questioning support, and richer citations. Tencent has integrated DeepSeek into various platforms, including Tence

Read more

Elon Musk: Sam Altman Could Drop OpenAI Bid If He Sticks to Non-Profit Mission

2025-02-14

Elon Musk's $97.4 billion bid to acquire OpenAI's non-profit assets faces opposition from the company's board. Musk wants to preserve OpenAI's non-profit mission, challenging CEO Sam Altman's plan to transfer control to investors like Microsoft. The conflict stems from a long-standing dispute since 2015, with Altman seeking to separate from the non-profit entity while Musk aims to complicate these efforts through legal action. OpenAI must finalize its restructuring within two years or risk conve

Read more

Alibaba DAMO Academy Open-Sources VideoLLaMA3: 7B Model Reaches New Heights in Video Understanding

2025-02-14

Damo Academy's VideoLLaMA 3 is a video-language model with 7 billion parameters that excels in video understanding tasks, including temporal reasoning and long-video comprehension. Built on an image-centric architecture, it outperforms open-source models of similar scale using just 3 million video-text datasets. Key innovations include Arbitrary Resolution Vision Tokenization (AVT) and Differential Frame Pruner (DiffFP), enabling efficient processing of diverse resolutions and reducing redundant

Read more

Kunlun Tech Unveils Matrix-Zero World Model, Driving Spatial Intelligence Innovation

2025-02-14

Kunlun Tech launched the Matrix-Zero World Model, a breakthrough in spatial intelligence technology. It features two core sub-models: one for generating realistic 3D scenes with dynamic effects, and another for creating interactive videos based on user input. This AI-driven model enhances digital content creation efficiency, reshaping industries like gaming, film, and embodied intelligence. Set for release in April, it will boost AI-powered game and video production, potentially revolutionizing

Read more

AI-Powered Humanoid Robotics Company Apptronik Raises $3.5 Billion

2025-02-14

Apptronik, a company developing AI-powered humanoid robots, secured $3.5 billion in Series A funding led by B Capital and Capital Factory, with Google participating. The funds will advance their next-gen robot, Apollo, designed for industrial applications like warehouses and manufacturing. Apollo aims to collaborate with humans, performing tasks such as heavy lifting and inspections. Apptronik collaborates with Google's DeepMind and partners with Mercedes-Benz and GXO Logistics. They plan to exp

Read more

Arm Teams Up with Meta to Launch Self-Developed Chips This Year

2025-02-14

Arm is set to launch its own chip development project targeting server CPUs for data centers, with Meta as a key client. The customizable chips will be outsourced for production, and the first in-house chip may debut this summer. This shift marks a move from licensing designs to competing directly with current clients like Apple and NVIDIA.

Read more

Elon Musk Claims His New AI Chatbot Is "Scary Smart" – Set to Launch in Weeks

2025-02-14

Elon Musk revealed at the Dubai World Government Summit that xAI's Grok-3 chatbot may launch within weeks. Described as "scarily smart," it surpasses other AI models in testing, using synthetic data and self-correction for logical consistency. Trained on 200 million GPU hours, Grok-3 integrates with X for real-time data and includes an unrestrained mode. Though not fully ready, Musk suggests a release could occur in one or two weeks. The AI landscape is shifting toward reasoning models, with xAI

Read more

Anthropic Expected to Release New AI Model Within Weeks

2025-02-14

AI startup Anthropic is set to launch its next-gen hybrid AI model soon, capable of toggling between deep reasoning and quick responses. The model will include a cost-saving sliding scale feature due to the high computational demands of deep reasoning. It outperforms OpenAI's o3-mini-high in some programming tasks and large codebase analysis. CEO Dario Amodei mentioned the model's unique inference capabilities during an interview.

Read more

Alibaba Partners with Apple to Bring AI Features to iPhone for China Market

2025-02-14

Alibaba and Apple have partnered to integrate AI into iPhones for the Chinese market, aiming to boost sales amid declining figures. Apple's iPhone sales in China dropped 11% year-over-year, with intensified local competition from vivo and Huawei reducing its market share. The collaboration leverages Alibaba's AI tech to enhance performance, addressing limitations of Apple's own generative AI solutions. Regulatory approval remains pending, while trade tensions and tariffs pose ongoing risks to Ap

Read more

New Open-Source AI Model Matches DeepSeek in Performance with Significantly Less Training Data

2025-02-14

An international team unveiled OpenThinker-32B, an open-source AI model that matches or surpasses DeepSeek in reasoning tasks. It achieved 90.6% accuracy on MATH500 and outperformed DeepSeek in GPQA-Diamond and LCBv2 benchmarks, despite using fewer training samples. Built on Alibaba's Qwen2.5, it offers a 16,000-token context window and trains efficiently with H100 GPUs. Unlike DeepSeek, OpenThinker provides full transparency, enabling reproducibility and community-driven improvements. Available

Read more