Welcome back to AI Hungry, where the latest advancements in artificial intelligence are always on our radar. Microsoft has been pushing the envelope with new Azure AI capabilities, integrating cutting-edge models and expanding their catalog. With the addition of models like Phi 2 and the integration of Meta's Llama 2, the AI landscape is evolving rapidly.
Dive into our latest articles to discover how these innovations are redefining the potential of AI and what it means for the future of technology.
Microsoft Unveils New Azure AI Capabilities and Expands Model Catalog
Microsoft has introduced new features to its Azure AI platform, including the integration of Meta's Llama 2 in its Models as a Service. Additionally, they have launched a preview of GPT-4 Turbo with Vision capabilities and expanded their model catalog with six new options such as Phi 2, Orca 2, and Mixtral 8x7b.
Fine-tuning services for GPT-3.5 Turbo and other models are now widely available, marking a significant enhancement in the AI offerings provided by Microsoft. These advancements are set to further transform the interaction between humans and computers.
Microsoft Unveils Phi-2: A Compact AI Model Surpassing Larger Counterparts
Microsoft has introduced Phi-2, a new small open source language model that boasts significant performance despite its modest size of 2.7 billion parameters. This model is part of the Phi project, which aims to create powerful yet small language models. Phi-2 has demonstrated superior capabilities, outperforming models up to 25 times its size, including Meta's Llama 2-7b and Google's Gemini Nano in various benchmark tests. It excels in areas such as commonsense reasoning and Python coding, building upon the knowledge and capabilities of its predecessors, Phi-1 and Phi-1.5.
Phi-2 is available for non-commercial research purposes through Azure AI Studio and Hugging Face, aligning with Microsoft's focus on responsible AI development. The model is noted for its safer outputs, showing reduced toxicity and bias without the need for reinforcement learning from human feedback. Microsoft's approach to training Phi-2 involved using high-quality data sources and synthetic datasets, which has led to its enhanced performance and efficiency in training.
🔌 Intel Launches 5th Gen Xeon CPUs with AI Acceleration and Enhanced Efficiency. Intel's new 5th Generation Xeon Scalable processors, dubbed Emerald Rapids, offer up to 42% faster AI inference, 21% more general compute performance, and 36% higher performance per watt than previous models, promising major data center efficiency and cost savings. (Link)
🌐 Albania Embraces ChatGPT for EU Application Process Amid Global AI Integration in Governance. Albania is partnering with OpenAI to use ChatGPT for translating EU legal documents, a move that could either speed up or hinder its EU accession. Meanwhile, AI is being adopted globally in government for tasks like fighting corruption and enhancing services. (Link)
🖨️ MIT Unveils Groundbreaking 3D Printer with Machine Vision for Faster, More Complex Creations. MIT researchers have developed a new 3D printer that uses machine vision to work with a broader range of materials, enabling faster production and more complex objects like robotic hands and heart-like pumps. (Link)
🧠 Google Deepmind's FunSearch Breakthrough in Mathematical Problem-Solving. Google Deepmind's FunSearch, combining a language model with an evolutionary algorithm, discovered a new solution to the 'cap set problem'. It also outperformed humans in solving the bin-packing problem, showcasing potential for broader applications. (Link)
🚀 Google Launches MedLM AI Models for Healthcare Sector. Google introduces MedLM, AI models tailored for healthcare, boasting 85% accuracy akin to medical specialists. Available in the US and select markets, these tools support tasks in hospitals, drug development, and more, with plans to enhance them with Gemini-based models soon. (Link)
🖥 China Unveils Tianhe Xingyi Supercomputer, Withdraws from Top500 List. China's National Supercomputer Center in Guangzhou has introduced Tianhe Xingyi, a CPU-based supercomputer for high-performance tasks. Despite its capabilities, it won't feature on the Top500 list as China has ceased participation. (Link)
