German AI robotics startup Sereact GmbH raised €25M to enhance AI models and robotic platforms. Investors include Creandum, Point Nine, and Air Street Capital. Their technology enables robots to interact with objects in real-time without extensive training, aiming to support more complex tasks beyond logistics and manufacturing.

OpenAI CEO Sam Altman urges calm on AGI, contradicting his earlier predictions. He dismisses imminent deployment and highlights the need to lower expectations. Experts question AGI's true capabilities, noting it hasn't reached consciousness yet. OpenAI remains tight-lipped on o3 model details.

Cybersecurity startup Tenex.AI launches AI-powered MDR service for Google Cloud with three tiers including 24/7 monitoring and automated detection maintenance. Backed by Andreessen Horowitz, the company plans to expand into compliance, identity governance, and vulnerability management.

Researchers restored ELIZA, the first electronic chatbot code from the 1960s by Joseph Weizenbaum. Found in MIT archives, the code was cleaned and adapted to run in a compatible environment. Despite known bugs, it functions well as a request/response tool, showcasing early AI conversational capabilities.

DouBao app updated to version 7.2.0 with Real-Time Voice Mega Model integrating ASR, NLP, and TTS for improved voice conversation. Users rate it higher than GPT-4o in naturalness and emotional expression. The model uses pre-training and reinforcement learning for enhanced safety and quality, featuring real-time internet connectivity and low-latency interaction.

World's first large-scale seismic model "Diti" with over 100 million parameters launched its third test version at Chengdu Supercomputing Center analyzing Dingri Tibet 6.8 magnitude quake data effectively. Public release planned for 2025 enabling business and research use. Developed by China Earthquake Administration, Tsinghua University, and Chengdu Center, it detected 452 pre-quake events and 429 aftershocks within 27 hours demonstrating significant potential in advancing seismology.

Researchers from the University of Hong Kong and Kuaishou Technology introduced GameFactory, a framework leveraging pre-trained video diffusion models and a multi-stage training strategy to generate diverse game environments. It excels in scene generalization and action control, outperforming existing methods in both discrete and continuous control signals, showcasing strong potential for generative game engines.

ByteDance launches Trae, an AI-integrated IDE with Chinese interface featuring Claude 3.5 and GPT-40 models. Trae offers Builder and Chat modes for project initiation and code-related queries. It supports automatic code completion, optimization, and refactoring. Users can import configurations from VSCode or Cursor and access Chat via shortcuts. Trae does not access internet resources directly and recommends setting up a virtual environment for project generation. Command-line tools are also sup

OpenAI is developing a new super AI agent expected to debut on January 30, aiming to solve complex PhD-level problems. The agent can perform tasks with minimal human intervention and adapt to various situations. OpenAI CEO Sam Altman plans to discuss AI's economic impact with US officials. However, some employees express concerns about potential risks from rapid AI advancements. The upcoming tool, rumored as "Operator", promises enhanced user convenience by executing multiple complex tasks.

Researchers introduced HiCo, a hierarchical video token compression method, and VideoChat-Flash, a system for long-context video understanding. HiCo compresses tokens to expand context windows while retaining essential data, improving computational efficiency. VideoChat-Flash uses a multi-stage learning scheme and extensive datasets to enhance visual perception and sequence parallelism. The model excels in short and long video benchmarks, achieving up to 99.1% accuracy in the NIAH task and setti

Microsoft releases AutoGen v0.4 with async event-driven architecture, enhancing multi-agent systems for enterprise use by improving task execution speed and resource efficiency. It integrates seamlessly with Azure, offering customization and pre-built apps via Copilot Studio. Compared to LangChain and CrewAI, AutoGen stands out for its Azure support and scalability, making it ideal for enterprises already within the Microsoft ecosystem. The focus is shifting towards practical applications, empha

Humanoid robots currently focus on specific tasks like manufacturing but will expand to broader applications as technology matures. Challenges include technical issues and public trust, with efforts being made to make robots more approachable and ethical. By 2050, an estimated 63 million humanoid robots could be in use in the US, potentially reshaping societal norms and addressing labor shortages. Data transparency and ethical behavior are crucial for their acceptance.

Metropolis protocol integrates AI agents with blockchain via dedicated chainlets, facilitating seamless cross-chain communication and fostering new services. This initiative by Saga, AI16z, Virtuals, and Wayfinder enables specialized AI functionalities and secure, isolated environments for smart contracts. Despite risks like centralization and security vulnerabilities, the AI-blockchain market is booming, with a $77.6B market cap, attracting significant investment. Metropolis stands out by offer

Scott Woody, CEO of Metronome, aims to revolutionize billing systems by offering flexible, real-time solutions for SaaS and AI enterprises. Metronome's platform allows companies to monitor and adjust pricing dynamically, enhancing user experience and cost control in the emerging AI era. Clients include OpenAI, Anthropic, Nvidia, Databricks, and Snowflake. Metronome addresses technical challenges by ensuring instant billing updates and providing enhanced visibility into AI expenditures.

Instabase secures $100M in Series D funding led by Qatar Investment Authority, now valued at $1.24B. The platform handles unstructured data, automating extraction and standardization from various documents and supporting over 160 languages. It offers advanced features like natural language prompts and a chatbot for data interaction, aiming to enhance AI-driven data processing capabilities.

Google is positioning itself in the rapidly growing multimodal AI market by enhancing BigQuery to handle diverse data types, aiming to derive insights from unstructured data. Applications include optimizing staffing levels at fast-food chains and improving delivery routes for UPS. Multimodal AI integrates various data sources for advanced analytics, enabling targeted marketing and personalized customer experiences while accelerating application development. Early adoption focuses on internal use

Apple plans to introduce AI-powered email categorization and prioritization features in macOS 15.4 Mail app, enhancing efficiency. Similar features were previously implemented in iOS 18.2, showcasing Apple's expertise in smart email management. These updates aim to simplify handling large volumes of emails for users.

Microsoft tests AI-powered search in Windows 11 Insider builds, enabling natural language searches for local files using semantic indexing. This feature requires a Copilot Plus PC and operates offline. Initially limited to specific file formats and Windows settings, it will expand to cloud services like OneDrive. Supported languages include Chinese, English, French, German, Japanese, and Spanish. An AI writing tool assists with grammar correction.

OpenAI is launching o3-mini, a new AI model with enhanced reasoning capabilities for complex tasks in various fields. Set for release after safety tests, it follows the o1 model and precedes the more extensive o3. OpenAI also plans to release APIs and ChatGPT features concurrently. Meanwhile, Google unveiled Gemini 2.0 Flash Thinking Experimental, intensifying AI competition. OpenAI's advancements, including protein engineering models, highlight its leadership in AI technology.

Google's Titans and Sakana's Transformers Squared introduce innovative memory architectures enhancing AI adaptability and efficiency. Titans features multi-level memory systems for improved long-term retention and context handling, excelling in extensive sequence processing. Transformers Squared employs a dual-pass mechanism with dynamic expert blending for real-time task adaptation, reducing computational demands while maintaining high performance. Both models aim to surpass traditional transfo