Smart Mask
Highlights
1. Digital twins—advanced computer models that replicate real-world counterparts—have become increasingly sophisticated and valuable for businesses and scientific research.
2. Cerebras has launched an AI inference service that significantly outperforms traditional cloud AI services, like AWS, opening up new possibilities for agentic AI applications.
3. Gemini Advanced subscribers will soon be able to create custom Gems, while all users will gain access to enhanced image generation features powered by Imagen 3.
4. Cursor is an AI-powered coding tool that enables users with limited coding skills to quickly develop apps.
5. Researchers have developed a smart mask that analyzes biomarkers in exhaled breath condensate to monitor various health conditions.
6. The 2024 World Robot Conference in Beijing highlighted a wide range of commercial applications for humanoid robots.
Innovation Insights
1. AI Won’t Give You a New Sustainable Advantage (Harvard Business Review)
The article argues that while generative AI (gen AI) has the potential to transform business operations and create value by automating processes, generating innovative ideas, and improving customer interactions, it is unlikely to provide a sustainable competitive advantage on its own. Since gen AI tools are widely accessible, any early advantages gained by companies will quickly diminish as competitors adopt the technology. The real strategic advantage lies in leveraging gen AI to amplify existing, hard-to-replicate resources and capabilities unique to a company, rather than relying on the technology itself to create new, lasting competitive edges.
2. Digital twins (The Economist 1, 2, 3)
Digital twins, advanced computer models that replicate real-world counterparts, have become increasingly sophisticated due to advancements in data collection and analysis technologies. They are being used to customize medical treatments, predict changes in planetary processes, and optimize large-scale scientific experiments, making them an essential tool for modern scientific research. They will revolutionize various manufacturing and maintenance processes, from predicting engine problems in aircraft to potentially monitoring everyday consumer products like cars and appliances. Creating digital replicas of businesses will become easier, enabling firms to seamlessly manage operations, forecast outcomes, and continuously optimize processes, though this also increases vulnerability to cyber threats.
AI Innovations
1. AI inference
Cerebras launched an AI inference service claimed to be the world’s fastest, using its WSE-3 chips to deliver speeds up to 20 times faster than conventional cloud AI services, enabling new possibilities for “agentic” AI applications (ZDNet).
2. Video analysis
Alibaba’s new AI model, Qwen2-VL, enhances video and image analysis capabilities by processing videos longer than 20 minutes and supporting multiple languages, offering advanced functionalities such as real-time video comprehension and integration with third-party applications, with two smaller versions available as open-source (VentureBeat).
3. AI workflow
NVIDIA and its global partners have launched NIM Agent Blueprints, a set of customizable AI workflows designed to help enterprises quickly develop and deploy generative AI applications for various use cases such as customer service, drug discovery, and data extraction (NVIDIA).
4. Google
Google Meet has introduced an AI-powered “take notes for me” feature for select Workspace customers, automatically summarizing meeting discussions and generating a Google Doc with key points, which is attached to the calendar event and shared with participants (The Verge).
Gemini Advanced subscribers will soon be able to create custom Gems, and all users will have access to enhanced image generation features powered by the new Imagen 3 model (Google).
Google announced new experimental updates to its Gemini AI models, including improved versions of Gemini 1.5 Flash and Gemini 1.5 Pro, which boast significant performance gains in areas such as coding, complex prompts, and handling large multimodal inputs (VentureBeat).
Google Research’s GameNGen is the first neural model-powered game engine capable of real-time simulation and interaction with complex environments, achieving over 20 frames per second while running the classic game DOOM on a single TPU (GitHub).
5. OpenAI
OpenAI’s upcoming AI model, “Strawberry,” is expected to significantly advance AI capabilities, particularly in reasoning and problem-solving (Newsweek).
Arizona State University is integrating ChatGPT across over 200 projects in teaching, research, and operations to personalize learning, enhance educational outcomes, and advance research (OpenAI).
6. KAN
A new approach to building neural networks, called Kolmogorov-Arnold Networks (KANs), simplifies the internal structure of artificial neurons by moving some complexity outside the neurons, potentially making AI systems more understandable and easier to interpret (MIT Technology Review).
7. New chip
IBM’s new Spyre accelerator chip, unveiled at Hot Chips 2024, is designed to enhance enterprise AI capabilities on IBM Z systems by enabling greater scalability, efficiency, and security for AI workloads (IBM).
8. Small model
Pharia-1-LLM-7B is a new, multilingual foundation model family, available under an open license for non-commercial use, with two variants—Pharia-1-LLM-7B-control and Pharia-1-LLM-7B-control-aligned (Aleph Alpha).
9. Biomedical
The new AI model ActFound outperforms competing models in bioactivity prediction and offers a cost-effective alternative to traditional methods (South China Morning Post).
Researchers have developed an AI tool, AINU, that can detect cancer cells and early viral infections within cells at nanoscale precision using high-resolution imaging (Medical Express).
10. Coding tool
Cursor is an AI-powered coding tool that allows users, even those with minimal coding experience, to quickly build apps by converting simple text prompts into functional code, using models like Claude 3.5 Sonnet and GPT-4o, and integrating with a code editor similar to Visual Studio Code (Tom’s Guide).
11. Long context model
Magic has introduced ultra-long context models, capable of reasoning over 100 million tokens during inference, aiming to improve AI’s understanding and performance in software development by accessing vast amounts of context data (Magic).
Other Innovations
1. Smart mask
Researchers at Caltech have developed a smart mask that analyzes biomarkers in exhaled breath condensate to monitor health conditions like asthma and COPD in real-time, providing continuous, noninvasive health assessments (MIT Technology Review).
2. Robot
Shanghai-based startup Agibot has unveiled a series of humanoid robots, including the flagship Yuanzheng A2, to compete with Tesla’s Optimus in the advanced robotics market, highlighting its technological breakthroughs and cost-control advantages (Interesting Engineering).
At the 2024 World Robot Conference in Beijing, humanoid robots were the main attraction, with 27 models showcasing capabilities in industrial training and commercial applications, including demonstrations like playing the piano and washing clothes. The event featured diverse robots, such as delivery, medical, and quadruped robots (KrEurope).
A new system developed by the German Aerospace Center enables robots to sense and interpret human touch without artificial skin, enhancing intuitive human-robot interactions by using joint-based sensors to detect touch and communicate commands (MIT Technology Review).
3. 3D printing
Washington State University researchers developed an AI-based method using Bayesian Optimization to optimize 3D-printing settings, significantly enhancing the efficiency and precision of manufacturing intricate structures like organ models, flexible electronics, and wearable biosensors (Washington State University).
Researchers at the University of Notre Dame have developed a 3D printing method using a laser to shape glass, enabling the creation of precise, transparent, bubble-free glass structures for applications in optical, microfluidic, and photonic devices (IEEE Spectrum).
