Apple reassigns veteran Kim Vorrath to boost AI and Siri development. Focus on revamping Siri's architecture and improving models. Delayed iOS 18.3 and 18.4 AI features signal cautious rollout amid accuracy concerns. Apple prioritizes AI, potentially over Vision Pro.

OpenAI's StarGate Initiative, valued at up to $500B, faces funding uncertainty. Jointly launched with SoftBank and Oracle, it aims to build AI-optimized data centers. Elon Musk criticizes the plan, questioning SoftBank's finances. Meta plans $60-65B capex, focusing on AI and LLMs like Llama 4.

OpenAI's o3 model excelled in FrontierMath, but controversy arose as OpenAI funded and accessed test materials. Though contracts barred using these for training, experts doubt fairness. Similar issues plague AI benchmarks like MMLU, GSM8K, raising concerns on genuine model evaluation capabilities.

OpenAI's Operator, an AI agent for automating tasks like bookings and reservations, is now available to US Pro Plan users. Using GPT-4's NLP and visual capabilities, it streamlines daily activities but requires human oversight for complex tasks. Future updates aim for broader integration.

ByteDance plans $20.64B capex for AI, half overseas. Benefits NVIDIA, Cambricon, Huawei. Leasing strategy due to procurement restrictions. Leading in China with 15 AI apps like Duobao. Competes with Baidu, Tencent. Lower spend vs US firms. TikTok's US future uncertain.

Microsoft's SIGMA, a system-domain large language model featuring DiffQKV attention, excels in AI infrastructure tasks. With 6 trillion pre-training tokens, SIGMA outperforms GPT-4 by 52.5% on AIMICIUS benchmark, optimizing inference speed and memory use.

Anthropic's new Citations feature links AI responses to specific document content via API and Vertex AI. It improves recall accuracy by 15% and simplifies workflow. Supporting Claude 3.5, Sonnet, Haiku models, it reduces misinformation risk in legal, financial fields. Developers benefit from enhanced traceability.

Google's Gemini app introduces smart home controls enabling natural language commands for devices like lights and thermostats. Users can execute multiple commands simultaneously and check device statuses. Non-sensitive device control from lock screen and enhanced media playback features are included. The update integrates Google Home functions directly into Gemini for seamless smart home management.

Perplexity Corp launched an Android AI assistant capable of multimodal tasks like composing emails, setting reminders, and booking reservations. It integrates with apps such as Spotify and Uber but lacks support for some platforms. Samsung and Google also enhance their AI assistants.

ChatGPT experienced a brief outage affecting over 10,000 UK users Thursday. OpenAI resolved the issue by 1509 PT and is monitoring results. Despite inconvenience, ChatGPT remains popular with over 300M weekly users by Dec 2024. Paid tiers up to $200 monthly are available.

OpenAI launched Operator, an AI agent for web browsing and task execution. Using GPT-4's visual and reasoning skills, it interacts with GUIs, seeks user input for sensitive tasks, and avoids harmful content. Currently limited in handling complex interfaces. Future plans include wider user access.

Hugging Face releases SmolVLM-256M, a compact visual language model with 256M parameters, optimized for low-power devices and web browsers. It excels in document analysis and image description, outperforming larger models in benchmarks. A 500M version offers enhanced quality.

Software engineer Ananay Arora bought "OGOpenAI.com" for a nominal fee and redirected it to Chinese AI lab DeepSeek, known for releasing advanced open-source AI models. This act honors DeepSeek's contribution, akin to early OpenAI's openness. Last week, DeepSeek launched DeepSeek-R1, outperforming GPT-o1 in benchmarks. Unlike OpenAI, which rarely releases open versions of flagship models, DeepSeek offers free offline use, sparking discussions on AI openness and accessibility.

Adobe launches new Premiere Pro beta at Sundance Film Festival featuring AI powered media intelligence, improved search functionality, multilingual subtitle support, and enhanced performance with After Effects. New Cloud Camera feature enables easy upload from select Canon cameras. Adobe aims to streamline video editing processes for all filmmakers.

In Davos, Anthropic's CEO Dario Amodei discussed future AI innovations, revealing plans for new features in their chatbot and emphasizing AI's potential to surpass human intelligence. He highlighted the need for safe AI and regulatory frameworks, and announced upcoming web search functionality and advanced AI agents, positioning Anthropic ahead of competitors like OpenAI. Amodei predicts AI will exceed human capabilities by 2027 but warns of potential risks.

ByteDance's new AI agent UI-TARS excels in controlling computers and executing complex workflows through GUI understanding. Trained on 50B tokens, it outperforms GPT-4, Claude, and Gemini in various benchmarks. It handles multimodal inputs and operates across desktop, mobile, and web platforms with superior perception and reasoning capabilities, demonstrating significant potential in the competitive AI landscape.

Anthropic secures $1 billion from Google, raising total investment to $30 billion. It plans to raise $20 billion more, valuing the company at $60 billion. In 2025, Anthropic will launch new AI models, enhance chatbot Claude with voice and web browsing, and introduce Virtual Collaborator for workflow management and code compilation. Total funding to date: $147 billion.

Google updates AI assistant Gemini with enhanced cross-app capabilities and new features on Samsung Galaxy S25 series. Gemini now handles complex tasks, integrates with more apps, and replaces Bixby as the default assistant. Improvements include Circle to Search enhancements and better compatibility with assistive technologies. Live voice conversations and sharing features are exclusive to newer Galaxy and Pixel devices.

Samsung's Unpacked 2025 event showcased the Galaxy S25 series with AI features including content credentials for identifying AI-generated images and enhanced Google Gemini assistant. Accessibility improvements for Android include LE Bluetooth support and better Braille display compatibility. Side button functionality now quickly launches Google Gemini.

At the 2025 Samsung Unpacked, Google enhanced its Circle to Search feature, now supporting more visual content types like locations and trending images via AI Overviews without needing extra text. Users can circle items such as artworks for instant AI summaries and interact with numbers, emails, and URLs more easily on Android devices.