Gene editing
Highlights
1. SenseNova 5.0 is claimed to be better than GPT-4 in the majority of general usage scenarios, especially in Chinese-language usage scenes.
2. A startup called Profluent used generative AI to develop a new gene editor OpenCRISPR-1. This might be an example that we need to get ready for a “ChatGPT moment” for biotech.
3. Synthesia introduced digital avatars showing various emotions.
4. Ray-Ban smart glasses are now with multimodal AI, enabling multimodal interactions.
5. Reusing old products or components for new products is getting popular, such as in clothing.
6. The tech war between US and China will likely intensify, although it will be costly to the world.
Innovation Insights
1. We need to be ready for biotech’s ChatGPT moment (Time)
Eric Schmidt highlights the impending biotech revolution that could change numerous industries by programming biology like computers, drawing parallels to the digital revolution. This revolution, boosted by AI, could lead to groundbreaking developments in disease treatment, agriculture, and environmental management, indicating a pivotal “ChatGPT moment” for biotech.
2. Three ways tech leaders can help companies reduce emissions (MIT Sloan)
Tech leaders can reduce carbon emissions by integrating emissions-tracking into their business practices, particularly focusing on the IT sector through more energy-efficient data centers and cloud-based solutions. They are also vital in ensuring transparency and data-driven approaches in product development, helping to identify and implement emissions-reduction initiatives. Additionally, technology executives may offer digital emissions-tracking tools to external stakeholders like customers and suppliers.
3. Why people really buy upcycled products (Harvard Business Review)
Researchers analyzing consumer feedback from Etsy found that the primary appeal of upcycled products lies in their creativity, not just their sustainability. They recommend that companies interested in upcycling should incorporate materials from diverse industries to boost product appeal and facilitate cross-industry collaboration. Additionally, marketers should emphasize the creative reuse of materials in their campaigns to enhance consumer interest and appreciation for the innovative use of repurposed components.
4. The tech wars are about to enter a fiery new phase (The Economist)
The tech wars between the U.S. and China are intensifying as they battle over control of crucial technologies like chipmaking and green technology. This competition, marked by U.S. export controls and Chinese industrial advances, is reshaping global relationships and supply chains, potentially costing up to 1.2% of global GDP annually according to the IMF.
AI Innovations
1. AI models
SenseTime launched its SenseNova 5.0, a new large model, at its Tech Day in Shanghai, introducing a scalable “Cloud-To-Edge” full-stack product matrix suitable for various applications. The model is claimed to be better than GPT-4 in the majority of general usage scenarios, especially in enterprise applications and Chinese-language usage scenes (CNBC).
Microsoft has introduced Phi-3 Mini, the latest iteration in its series of compact AI models, featuring 3.8 billion parameters and optimized for efficiency on platforms like Azure, Hugging Face, and Ollama. This model is designed to perform tasks comparable to larger models but with lower computational demands (The Verge).
Apple has launched OpenELM, a series of small, open-source large language models designed for on-device operation. Released on Hugging Face, OpenELM comprises eight models, both pre-trained and instruction-tuned, ranging from 270 million to 3 billion parameters, tailored for efficient text generation without the need for cloud connectivity (VentureBeat).
2. LLM evaluation
It is challenging to use LLMs in healthcare, where mistakes can have serious consequences. The Open Medical-LLM Leaderboard aims to address these challenges by providing a standardized platform for evaluating and comparing the performance of LLMs on various medical tasks and datasets (Huggingface).
A new tool developed by Cleanlab, an AI startup born from a quantum computing lab at MIT, is designed to evaluate the reliability of outputs from large language models (LLMs) by scoring them between 0 and 1. This “Trustworthy Language Model” tool aims to help users distinguish credible responses from potentially misleading ones in high-stakes environments (MIT Technology Review).
3. AI device
Rabbit R1, a new AI device, offers a more appealing and user-friendly alternative to the Humane AI Pin, featuring a reliable screen, straightforward controls, and a vibrant display. Unlike the pricier Humane AI Pin, which also requires a subscription, the Rabbit R1 does not, and it includes playful, user-centric features that enhance its charm and practicality (Engadget).
4. Driving with natural language
LINGO-2 is an advanced Vision-Language-Action Model (VLAM) that integrates vision, language, and action to control and explain autonomous driving behaviors. This model, which is the first of its kind tested on public roads, allows for real-time driving commentary and decision-making insights. LINGO-2 not only enables more natural interactions with users through language prompts that influence driving decisions but also serves as a tool for deeper AI understanding and training (Wayve)
5. AI video
Synthesia has introduced a new series of AI-generated digital avatars, known as “Expressive Avatars”, capable of displaying a range of human emotions like happiness, sadness, and frustration (CNBC).
Adobe’s new generative AI model, VideoGigaGAN, can upscale blurry videos to up to eight times their original resolution, delivering higher quality and more detailed visuals without the common issues like flickering seen in other Video Super Resolution (VSR) methods. This advancement from Adobe Research provides a balance between the sharp, detailed enhancements of GAN models and the smooth, consistent output of other upscaling techniques (The Verge).
6. Model hosting
Amazon Web Services (AWS) is positioning itself as a key player in hosting and fine-tuning custom generative AI models through its new feature, Custom Model Import, in its enterprise AI suite, Bedrock. This feature allows companies to import their proprietary AI models into Bedrock, where they can benefit from robust infrastructure, tools for expansion and bias mitigation, and integration with existing workflows (TechCrunch).
7. Image creation
Adobe has launched the Firefly Image 3 Foundation Model, designed to enhance photorealistic image creation, available in beta on Photoshop and the Firefly web application. This new model introduces enhanced features for fast and productive ideation, including new auto-stylization capabilities and improved control through Structure and Style Reference tools, allowing for higher quality and more personalized image outputs (Adobe).
8. Text-to-360
Skybox AI, Blockade Labs’ AI art creation platform, can now create and download 360-degree 3D images in 8K resolution. This is useful for apps like the Apple Vision Pro mixed reality headset. Skybox AI is used in many industries, including entertainment, gaming, and education (VentureBeat).
Other Innovations
1. Gene editing
Generative AI has advanced to create blueprints for microscopic mechanisms capable of editing DNA. Profluent, a startup based in Berkeley, California, has developed a new gene editor called OpenCRISPR-1 using this technology, which follows the principles of CRISPR but innovates by learning from vast biological datasets to create new gene-editing tools. OpenCRISPR-1 is open-sourced and signifies a move toward more precise and adaptable gene editing technologies, potentially speeding up the development of tailored medical treatments (New York Times).
2. Kidney transplant
Lisa Pisano, a 54-year-old woman from New Jersey, recently became the second living recipient of a gene-edited pig kidney, which contains only a single genetic modification. Pisano’s transplant, part of ongoing efforts to address organ shortages, was performed by NYU Langone Health, and early indications suggest she is doing well without signs of organ rejection (MIT Technology Review).
3. Obesity drugs
The introduction of semaglutide, a synthetic drug mimicking an intestinal hormone, has changed the approach to managing obesity and type 2 diabetes, recently expanding to cardiovascular disease prevention. This drug, marketed as Ozempic and Wegovy, is now part of a broader spectrum of new medications that interact with the gut-brain axis to manage weight more effectively. As more drugs like semaglutide enter the pipeline, targeting various hormonal mechanisms, they promise enhanced weight management solutions (Inverse).
4. Robots
Sanctuary AI has introduced the seventh generation of its Phoenix robots, which feature significant enhancements such as a more human-like range of motion and improved visual and tactile perception, enhancing their ability to perform complex tasks more efficiently and over longer periods. These advancements, along with a 50x increase in the speed of task automation, mark a major step forward in the development of robots that closely mimic human capabilities and behaviors (PR Newswire).
5. Meta
The Ray-Ban Meta Smart Glasses have been updated to include multimodal AI, enhancing their functionality by enabling them to process and respond to various inputs like photos, audio, and text. This new feature allows users to perform tasks such as identifying plants, translating signs, and generating content descriptions through voice commands that interact with AI to deliver results audibly (The Verge).
Meta is expanding the availability of its Meta Horizon OS, previously exclusive to Meta Quest devices, to third-party hardware manufacturers, fostering a broader ecosystem of mixed reality devices. This initiative, which involves collaboration with leading global technology firms, aims to make it easier for developers to create and distribute mixed reality apps (Meta).
