Google Maps users have reported losing years of location history due to a technical issue. Google confirmed the data loss, stating that users with encrypted timeline backups can recover their data, while others without cloud backups are out of luck. The problem arose after Google moved timeline storage to local device storage, increasing the risk of permanent data loss for those without backups.

Tencent has launched HunYuan T1, a powerful reasoning model excelling in ultra-long document processing and complex topics like math and coding. Built on the Hybrid-Mamba-Transformer architecture, it reduces computational costs while enhancing performance in long-text reasoning and multi-task adaptability. Available via Tencent Cloud, it offers competitive pricing for API services.

Shengshu Technology collaborates with a top novel platform to apply AI in the anime short drama market. Vidu, its AI video platform, secures rights for seven popular web novels for concept films and adaptations. These novels, with over 10 billion reads, span various genres. AI animation reduces production time and costs, enabling wider IP reach. Vidu also launches a contest for users to create one-minute trailers using their technology, combining public votes and expert reviews for prizes and fu

Huawei's Pura X redefines foldable phones with a clamshell design, HarmonyOS 5, and AI integration. Featuring a 6.3-inch inner display, powerful camera, and no reliance on Android, it marks a significant shift for Huawei amid US sanctions. Starting at 7,499 yuan, the device showcases AI-driven features and premium build options, targeting both style and functionality while focusing initially on the Chinese market.

Instagram is testing an AI-generated comment suggestion feature, represented by a "starred pencil" icon. Users can choose or edit AI-generated comments tailored to post content. This aims to enhance user engagement across Meta platforms, including Facebook and Instagram. While it simplifies commenting, concerns arise about authenticity and potential spam. Meta also launched a limited AI chatbot in Europe and is developing a specialized AI training chip for improved efficiency.

GPT-4o Copilot, an advanced AI model for Visual Studio Code, offers improved code suggestions through training on 275,000 public repositories and supports over 30 programming languages. It integrates via the Copilot menu and enhances developer productivity with context-aware completions. Key features include next-step editing suggestions and broader IDE support, though it lacks prompt caching and local models for code merging. Switching models doesn't affect Copilot Chat or data policies, and us

At GTC 2025, 1X showcased NEO, a tendon-driven humanoid robot capable of smooth human-like movements. Weighing 30 kg, NEO will enter homes this year for data collection via hybrid operation, combining autonomous balance with remote-controlled tasks. CEO Bernt Børnich emphasized the importance of real-world learning and compared owning NEO to raising a puppy, highlighting the journey toward full autonomy while addressing privacy concerns and seeking early adopters.

Google is enhancing Gemini Live with AI features from Project Astra, enabling real-time screen reading and live video interpretation. These updates, rolling out to Google One AI Premium subscribers, allow Gemini to analyze visual content and answer related queries. This move solidifies Google's AI leadership as competitors like Amazon and Apple work on similar technologies. Samsung now uses Gemini as its default assistant over Bixby.

Cloudflare launched AI Maze, a tool to combat unauthorized data scraping for AI training. It redirects improper bot activities to fake AI-generated pages, wasting malicious actors' resources without affecting human visitors. This opt-in feature aims to detect new bot patterns and improve bot management. Website admins can enable it via the Cloudflare dashboard.

Stepwise Star introduces Step-Video-TI2V, an open-source image-to-video model. Built on the 30B-parameter Step-Video-T2V, it generates 5-second, 540P videos with 102 frames. Key features include adjustable motion intensity and camera movement control, enhancing dynamic range and stability. Optimized for anime tasks, it supports multi-size image inputs and special effects. Available on Huawei's Ascend platform via Modelers community.

OpenAI integrates new AI models for transcription and speech generation into its API, enhancing previous versions. These align with their "agent" vision, offering customizable voices and reducing inaccuracies. The text-to-speech model, "gpt-4o-mini-tts," provides realistic voices with controllable tones. New speech-to-text models, "gpt-4o-transcribe" and "gpt-4o-mini-transcribe," replace Whisper, improving accuracy and reducing hallucinations. However, transcription accuracy varies across langua

Google is upgrading Gmail's search with an optional AI-driven feature to improve email retrieval. The new system prioritizes relevance by considering factors like recent access frequency, user interactions, and frequent contacts. Rolled out globally for personal accounts, it complements the traditional chronological search without replacing it. Users can switch between both options as needed.

The Google Pixel 9A features the Tensor G4 chip but has reduced Gemini AI capabilities due to its 8GB RAM. It uses a smaller Gemini Nano 1.0 XXS model compared to higher-end models with 12-16GB RAM. Key AI features like Pixel Screenshots and Call Notes are unavailable, though audio summaries in Recorder remain functional. This contrasts with last year's Pixel 8 Pro-exclusive AI features, later extended to the Pixel 8 after feedback.

Nvidia's new RTX 5080 and RTX 5090 GPUs are hard to obtain, sparking "paper launch" criticisms. During GTC 2025, Nvidia will sell 1,000 units of each model at retail price via its Gear Store, but only attendees paying $1,145/day can purchase. Acquiring these GPUs remains expensive and difficult.

Anthropic's chatbot Claude now features web search capability, currently available to paid US users as a "feature preview." This integration allows Claude to provide up-to-date information with citations for verification. While complicating implementation due to the probabilistic nature of language models, this feature may expand to free users and more countries soon. Despite higher operational costs, Anthropic does not plan to increase user fees but might adjust pricing for new functionalities.

Apple reshuffles its Siri division as Mike Rockwell replaces John Giannandrea as head of AI and Siri. The move follows a delay in launching key AI features, with Apple postponing Siri's generative AI upgrade to next year. Tensions within the team highlight challenges in AI development, while Paula Midgley takes over Vision Pro leadership.

NVIDIA introduces the Llama Nemotron series, enhancing AI inference with improved math, coding, and reasoning capabilities. It offers a 20% precision boost and fivefold speed increase over other open-source models, enabling complex decision-making and cost savings. Major AI platforms like Accenture, Microsoft, and SAP are integrating it into their operations.

NVIDIA and Microsoft strengthen their partnership to advance AI with the integration of NVIDIA's Blackwell architecture into Azure. Key highlights include the launch of Azure ND GB200 V6 VMs featuring NVIDIA technology, incorporation of NVIDIA NIM into Azure AI Foundry for simplified generative AI deployment, and optimization of language models like Meta Llama. Additionally, new offerings such as NVIDIA Omniverse and Isaac Sim are added to the Azure Marketplace to support AI developers.

NVIDIA unveiled two compact personal AI supercomputers, DGX Spark and DGX Station, designed for developers and researchers. DGX Spark, priced at $3,000, is claimed to be the world's smallest supercomputer with 784 GB memory and supports AI-native applications. DGX Station offers 784 GB unified memory and advanced networking capabilities, ideal for large AI workloads. Both systems are built on NVIDIA's Blackwell Ultra platform and integrate with NVIDIA's AI development ecosystem.

Noam Brown from OpenAI discussed how AI reasoning models could have emerged earlier with the right methodologies. He highlighted the importance of human-like contemplation in AI systems and his work on Pluribus and OpenAI's o1 model, which use reasoning techniques for better accuracy. Brown acknowledged the challenges academia faces due to limited resources but suggested focusing on areas like model architecture or AI benchmarking, which currently lacks proper standards and doesn't require heavy