News
1. AI and image
Apple has introduced a new generative AI model for image editing, named MGIE, developed in collaboration with the University of California, Santa Barbara, which allows users to modify images through simple text prompts for tasks like cropping, resizing, flipping, and adding filters. This open-source model, available on GitHub and demonstrated on Hugging Face Spaces, represents a significant step for Apple in the generative AI space, enabling intuitive, language-based photo editing without the need for traditional photo editing software (The Verge).
Google Lookout, an app designed to assist the visually impaired and blind by using AI to describe their surroundings via phone cameras, has introduced a new feature called Image Q+A. This feature allows users to ask questions about images and receive detailed descriptions, enhancing their understanding of visual content (ZDNet).
2. Open-source AI
Smaug-72B, a new open-source language model developed by Abacus AI, has been declared the best in the world according to Hugging Face’s rankings, outperforming other advanced models like GPT-3.5 and Mistral Medium in popular benchmarks. This model, a fine-tuned version of Qwen-72B, marks a significant milestone in open-source AI, showcasing the potential to rival proprietary technologies and democratize AI innovation (Venture Beat).
Mistral’s CEO confirmed the leak of a new open-source AI model named “miqu-1-70b,” which has shown performance nearing that of OpenAI’s GPT-4, stirring excitement within the open-source AI community. This model, initially shared on HuggingFace and discussed across various online platforms, represents a significant step for open-source AI, potentially challenging the dominance of proprietary models like GPT-4 with its high benchmark scores and the promise of further improvements (Venture Beat).
3. Deciphering ancient text
Using AI, researchers have successfully deciphered text from a 2,000-year-old Herculaneum scroll, revealing philosophical musings on music, pleasure, and capers, marking a significant breakthrough in the field of archaeology. This achievement, part of the Vesuvius Challenge, demonstrates the potential of AI in uncovering ancient texts, offering insights into previously inaccessible aspects of ancient Greek philosophy and setting the stage for further discoveries within the remaining Herculaneum scrolls (Nature).
4. Gemini
Google has rebranded Bard as Gemini and introduced Gemini Ultra, its most advanced large language model, available through a new $20 Google One AI Premium plan that includes 2TB of storage and access to Gemini in Google Workspace apps. Gemini Ultra aims to set the state of the art across various benchmarks and will be accessible in more than 150 countries in English, with plans to add more languages (TechCrunch).
5. AI assistant
Hugging Face has launched a new, free product called Hugging Chat Assistants, allowing users to create their own customized AI chatbots, rivaling OpenAI’s custom GPTs which require a paid subscription. Unlike OpenAI’s offerings, Hugging Chat Assistants are powered by a variety of open source large language models, offering a broad selection for backend intelligence, and emphasizing Hugging Face’s commitment to open source AI development (Venture Beat).
The Limbic chatbot, designed to screen individuals seeking mental-health assistance, significantly increased the number of referrals to England’s National Health Service (NHS) mental-health services, particularly among minority communities, according to new research. This AI-driven tool not only facilitated access to care but also improved the efficiency and quality of patient assessments without increasing waiting times, showcasing the potential of AI to address the supply-demand imbalance in mental health care and promote equity and inclusion (MIT Tech Review).
6. AI safety
The Biden administration announced that over 200 entities, including leading AI companies like OpenAI, Google, and Microsoft, have joined a new U.S. consortium aimed at supporting the safe development and deployment of generative AI. This consortium, named the U.S. AI Safety Institute Consortium (AISIC), is part of an effort to mitigate risks and harness the potential of artificial intelligence, focusing on priority actions such as developing guidelines for red-teaming, risk management, and watermarking synthetic content (Reuters).
7. AI agent
Arc is developing an AI agent designed to navigate the internet for users, aiming to streamline online browsing by conducting searches, comparisons, and summarizing information across various websites. This innovative tool seeks to enhance productivity and efficiency by automating the process of gathering and synthesizing web content, potentially transforming how users interact with the internet (TechCrunch).
8. Robotics
1X aims to provide an abundant supply of physical labor through safe, intelligent androids designed with human-like hardware for versatility, focusing on learning motor behaviors from vision using neural networks for autonomy. They have successfully deployed these androids for tasks like patrolling, showcasing their capabilities in a video where all actions are autonomously controlled by a vision-based neural network, demonstrating a significant advancement in training robots for a variety of tasks without traditional programming (1X).
9. Neurotech
Elemind, an AI-powered neurotech startup, has emerged from stealth with backing from high-profile investors including Jeff Bezos and Bill Gates, aiming to pioneer “electric medicine” through noninvasive wearable devices that deliver tailored neurostimulation based on real-time brain wave analysis. The company, founded by scientists from MIT, Stanford, and Harvard, focuses on offering a natural treatment alternative for neurological conditions such as insomnia, essential tremor, and memory loss, avoiding the side effects associated with pharmaceuticals (Fierce Biotech).
10. Wearable
The National University of Singapore has developed an experimental wearable AI device named AiSee, designed to assist visually impaired individuals by recognizing objects they hold and describing them via a computer-generated voice. This device, worn around the neck like headphones, uses a 13-megapixel camera to take photos of objects, which are then analyzed by AI algorithms to provide the user with details about the object’s size, shape, color, and any visible text (PC Gamer).
Brilliant Labs has announced the launch of Frame, touted as the world’s first glasses with an integrated multimodal AI assistant named “Noa,” designed to interact with the physical world through AI, including visual queries and search capabilities. Unlike other smart glasses, Frame focuses on providing an always-on AI experience without storing images or videos, aiming for a blend of technology and fashion with a retro design, and is available for preorder at $349 (ZDNet).
11. Energy storage
An abandoned mine in Finland is being repurposed into a giant gravity battery to store renewable energy, utilizing the mine’s deep shafts to lift a heavy weight with excess energy and release it to generate electricity during low production periods. This innovative project, developed by Scottish firm Gravitricity, aims to demonstrate the potential of using disused mines for reliable, long-life energy storage, offering a sustainable solution to capture and release energy as needed (Independent).
12. High-speed train
China Aerospace Science and Industry Corporation (CASIC), China’s largest missile manufacturer, has achieved a new milestone in hyperloop technology by clocking a superconducting maglev vehicle at over 623 km/h (387 mph) in a low-vacuum tube, setting a record for the fastest speed ever for such a vehicle. This development is part of CASIC’s efforts to create a 1,000 km/h high-speed maglev train running in a low vacuum pipeline, showcasing China’s commitment to advancing cutting-edge transportation technologies despite the high costs and practical challenges associated with hyperloop systems (New Atlas).
13. Directed evolution
Researchers at Cambridge have developed a new E. coli strain with an “orthogonal” DNA replication system that allows for the rapid evolution of genes by generating many random mutations without affecting the bacterium’s essential genes for survival. This system, inspired by a natural mechanism in yeast and enhanced through engineering, enables the accelerated evolution of specific genes, such as those for antibiotic resistance or green fluorescent protein, demonstrating a significant advancement in the field of genetic research and potential applications in medicine, research, and industry (Ars Technica).
14. Bioluminescent houseplant
Consumers in the United States can now pre-order a genetically engineered petunia that glows continuously in the dark, thanks to biotechnology firm Light Bio. This innovation, utilizing genes from the bioluminescent mushroom Neonothopanus nambi, allows the petunia to emit a faint green light without the need for special food or light, marking a significant advancement in the field of bioluminescent plants for decorative purposes (Nature).
15. Separating chiral molecules
Researchers have developed a method using mass spectrometry to separate chiral molecules, which are mirror images of each other but cannot be superimposed, potentially simplifying a complex part of drug discovery. This technique, demonstrated by Zheng Ouyang’s team at Tsinghua University, can distinguish between different forms of molecules like enantiomers, which have identical atoms but distinct spatial arrangements, offering significant implications for drug development and quality control (Nature).
Articles
1. How virtual power plants are shaping tomorrow’s energy system (MIT Tech Review)
Virtual Power Plants (VPPs) are transforming the energy landscape by aggregating distributed energy resources like solar panels and electric vehicle chargers to balance supply and demand, thereby reducing reliance on traditional power plants and aiding in the transition to clean energy. These systems, which operate without a central physical facility, offer unique advantages such as real-time energy use shaping and increased grid resilience, making them a crucial component in modernizing energy systems and reducing carbon footprints.
2. The risks of US-China decoupling (Financial Times)
The US’s strategy to restrict China’s access to high-end semiconductors, aimed at halting America’s role in China’s military modernization, may inadvertently accelerate China’s own technological advancements in the semiconductor industry. Despite the intention to limit the dual-use technology for military growth, evidence suggests that Chinese companies, such as SMIC and Huawei, are rapidly catching up by developing high-end products and potentially shortening the timeline for achieving technological parity with leading US firms.
3. Investors pile into psychedelic drug start-ups tackling mental health (Financial Times)
Biotechnology startups focusing on psychedelic drugs for mental health treatment are attracting significant investment, with sovereign wealth funds like Temasek and Mubadala entering discussions to fund these ventures. The sector, revitalized by promising clinical data and potential regulatory approvals, raised at least $163 million across five deals in January, marking a surge in interest as these substances, including MDMA, psilocybin mushrooms, and 5-MeO-DMT, show potential in treating mental health disorders.
4. Synthesizing and identifying potential biomarkers to explore uncharted biochemistry (Nature)
The study introduces a reverse metabolomics approach, leveraging public metabolomics data repositories to identify previously unknown metabolites by synthesizing thousands of small compounds using combinatorial chemistry and matching their molecular fingerprints against large databases. This innovative method has led to the discovery of new bile acid conjugates as potential biomarkers for inflammatory bowel disease (IBD), showcasing the potential to uncover novel biomarkers and understand unexplored biochemical pathways in human health and disease.
