AI Agents
Highlights
1. A new article discusses how companies can create adaptable products that evolve to meet consumers’ changing needs over time.
2. Nvidia released Llama-3.1-Nemotron-70B-Instruct, which outperforms GPT-4 and Claude 3.5 in benchmarks.
3. OpenAI’s new product “Swarm” enables multiple AI agents to autonomously collaborate and orchestrate actions using natural language.
4. Mistral AI has introduced two new state-of-the-art edge models, Ministral 3B and Ministral 8B, designed for efficient on-device computing.
5. Researchers at the University of Toronto have developed a flexible robotic system that uses computer vision and a language model to automate chemistry experiments.
6. Moflin is a fuzzy, limbless pet robot designed as a mental health companion.
7. SpaceX rocket booster successfully landed after test launch, caught by ‘chopsticks’ mechanism.
Innovation Insights
1. Design products that won’t become obsolete (Harvard Business Review)
The article discusses how companies can create adaptable products, referred to as “products that grow,” that evolve to meet consumers’ changing needs over time. These products, which can include items like software-updatable hardware, modular devices, and configurable tools, extend the useful life of offerings and enhance customer value. This approach creates opportunities for ongoing innovation, customer engagement, and competitive advantage while aligning with environmental and social responsibility goals.
2. Competitive advantage in the age of AI (California Management Review)
In the age of generative AI, companies must rethink their competitive advantage by focusing on six key areas: proprietary data, a strong digital core, rapid learning, deep capability reinvention, external partnerships, and building trust. Generative AI is reshaping traditional sources of advantage by commoditizing previous differentiators and making advanced capabilities more accessible. To stay competitive, businesses must not only adopt AI but also innovate through end-to-end workflow transformation, develop ecosystems, and establish responsible AI practices.
AI Innovations
1. Nvidia
Nvidia released a new AI model, Llama-3.1-Nemotron-70B-Instruct, which outperforms OpenAI’s GPT-4 and Anthropic’s Claude 3.5 in benchmarks (VentureBeat).
2. Microsoft
Microsoft Research and Tsinghua University introduced the Differential Transformer, a new LLM architecture that improves attention accuracy by filtering out noise and amplifying relevant context, leading to better performance in long-context tasks like information retrieval and in-context learning (VentureBeat).
3. Anthropic
Anthropic has announced a major update to its Claude AI apps, including the launch of an iPad app and new features like the ability to search past chats and customize instructions for more personalized user experiences across devices (Tom’s Guide).
4. OpenAI
OpenAI has launched a preview version of a dedicated ChatGPT app for Windows, available to select users, allowing features like file and photo uploads, document summarization, and image creation with DALL-E 3, though some functions like voice support and GPT Store integrations are not yet available (TechCrunch).
OpenAI’s new product “Swarm” introduces agentic AI, enabling multiple AI agents to autonomously perform tasks, collaborate, and orchestrate actions using natural language, marking a significant step in the development of multi-agent AI systems for real-world applications (Forbes).
5. Meta
Meta researchers, in collaboration with UC Berkeley and NYU, developed a technique called “Thought Preference Optimization” (TPO) to improve AI models’ reasoning by having them generate thought steps before answering, leading to better results across a variety of tasks, except for specialized ones like math (The Decoder).
6. Amazon
Amazon Ads has launched new AI tools, including AI creative studio and Audio generator, to help advertisers easily create, update, and enhance ad content across formats like images, video, and audio, improving engagement with customers (Amazon).
7. Alphabet
NotebookLM has introduced customizable Audio Overviews, allowing users to guide AI hosts’ focus and expertise level, and launched a NotebookLM Business pilot program offering enhanced features for businesses, universities, and organizations (Google).
Google Shopping is being transformed with AI-powered personalization, helping users find tailored products, research prices, and discover deals more easily through a revamped shopping experience (Google).
An AI-powered tool developed by Google DeepMind can help groups with differing views find common ground by synthesizing their opinions into fair, clear summaries, often preferred over human mediators’ efforts, and could improve democratic deliberation processes (Nature).
8. Perplexity
Perplexity has introduced a new Internal Knowledge Search feature, allowing enterprise users to search both internal uploaded files and the web from a single platform, enhancing productivity and combining insights from multiple sources (VentureBeat).
9. Small models
Mistral AI has introduced two new state-of-the-art edge models, Ministral 3B and Ministral 8B, designed for efficient on-device computing with enhanced knowledge, reasoning, and task-handling capabilities, setting a new standard in the sub-10B category for various privacy-first applications (Mistral AI).
Zyphra has launched Zamba2-7B, a state-of-the-art, efficient language model that outperforms leading 7B models like Mistral, Gemma, and Llama3 in both quality and performance, making it ideal for on-device and enterprise applications (Zyphra).
10. Adobe
At Adobe Max 2024, Adobe announced several updates, including the public beta of its first AI video generation model in Premiere Pro, new AI-powered tools for Photoshop, Illustrator, and InDesign, and showcased experimental features like Project Clean Machine, Scenic, and Know How for future creative enhancements (Forbes).
11. Pika labs
Pika Labs has expanded its AI video model, Pika 1.5, to include new special effects like crumbling, dissolving, and disappearing video subjects, further setting it apart from competitors in the evolving AI content generation space (VentureBeat).
12. Zoom
Zoom is introducing AI avatars that can mimic your appearance and voice to send short messages on your behalf, with the feature set to launch next year as part of a custom AI Companion add-on (The Verge).
13. Music
Team Suno has introduced a new feature, Suno Scenes, allowing users to create unique songs from their favorite photos and videos, expanding creative storytelling through visual content (Suno).
14. Diffusion forcing
MIT researchers developed a new method called “Diffusion Forcing,” which combines next-token prediction and video diffusion models to help AI agents and robots make flexible plans, sort noisy data, and perform complex tasks like video generation and manipulation in dynamic environments (MIT News).
15. AI and science
Archetype AI’s new model, Newton, learns complex physics principles directly from raw sensor data without human guidance (VentureBeat).
Meta is releasing a massive open-source data set and models, called Open Materials 2024, to help accelerate AI-driven materials discovery, providing researchers with much-needed data to simulate new materials for innovations like better batteries and sustainable fuels, which were previously hindered by costly, proprietary data sets (MIT Technology Review).
Researchers used AI to discover 70,500 previously unknown RNA viruses, many of which are found in extreme environments, revealing the potential of AI in exploring the vast “dark matter” of the RNA virus universe (Nature).
Other Innovations
1. Space travel
In a week that would normally focus on the launch of the Europa Clipper to explore oceanic life on a distant moon, SpaceX’s successful test flight of its massive Starship rocket, which aims to revolutionize space travel and enable Mars colonization, stole the spotlight and signaled a new era of ambitious space exploration and competition (The Economist).
2. Robots
Researchers at the University of Toronto have developed Organa, a flexible robotic system that uses computer vision and a language model to perform chemistry experiments based on verbal cues, automating complex lab tasks, analyzing data, and providing feedback to free up scientists for higher-level work (MIT Technology Review).
Ai-Da, a humanoid robot, is set to become the first of its kind to have its artwork auctioned at Sotheby’s, with its painting “AI God,” an abstract portrait of Alan Turing, expected to sell for over $120,000 (Popular Science).
The AI-powered Moflin is a fuzzy, limbless pet robot designed as a mental health companion that adapts its mood to its owner’s interactions, priced at $398 with preorders starting in Japan (Mashable).
3. Iris scanning
Worldcoin, now rebranded as “World,” unveiled new iris-scanning technology and tools to scale its human-verification blockchain network, while distancing its mission from cryptocurrency and emphasizing its goal to address AI-related challenges such as online identity verification and deepfakes (TechCrunch).
4. Dreaming
Bay Area startup REMspace claims to have developed technology enabling two-way communication between people through lucid dreaming (ABC 7 News).
