Scent Teleportation
Highlights
1. Osmo has developed an AI-powered “scent teleportation” system that captures and recreates scents remotely.
2. Magentic-One, a newly introduced multi-agent AI system from Microsoft AutoGen, enables flexible handling of open-ended web and file-based tasks.
3. NVIDIA’s new AI Blueprint enables developers to create visual AI agents that can analyze video and image data.
4. Claude AI’s new “Visual PDFs” feature allows users to analyze text, images, charts, and tables within PDFs.
5. Tencent has launched Hunyuan-large, an open-source MoE language model that excels in multilingual NLP, code processing, and mathematical tasks.
6. KLING now features “Custom Models,” allowing users to train unique video characters with consistent appearances across scenes by uploading video clips.
Innovation Insights
1. How ChatGPT search paves the way for AI agents (MIT Technology Review)
OpenAI is advancing towards developing AI agents that can perform complex tasks autonomously, such as booking flights, by introducing features like ChatGPT search and a Realtime API for voice integration. The two main challenges are improving the AI’s reasoning abilities and enhancing its ability to connect and interact with external tools, allowing for real-time information retrieval and action-taking. The o1 model, which includes a “chain of thought” feature, aims to enhance reasoning by breaking down tasks, although true reasoning remains a work-in-progress. Expanding use cases beyond coding and science to fields like law and economics is part of OpenAI’s roadmap.
2. Why your company is struggling to scale up generative AI (The Economist)
While 39% of Americans now use AI, only 5% of businesses apply it in production, with many in “pilotitis,” experimenting but hesitant to scale up due to legal, regulatory, and reputational risks, as well as uncertain ROI. Companies face challenges with messy, decentralized data and legacy IT systems, which complicate AI integration and can lead to errors. Additionally, AI adoption is slowed by a shortage of skilled talent.
3. Slack CEO: How to roll out artificial intelligence internally (Ideas Made to Matter)
Slack CEO Denise Dresser highlights strategies for integrating AI effectively within organizations. She identifies five AI user archetypes—maximalists, undergrounds, rebels, superfans, and observers—and emphasizes tailoring engagement to meet each group’s needs. Dresser advises leaders to prioritize clear, practical use cases, particularly for repetitive tasks, to ensure AI adoption feels beneficial and manageable. Additionally, she notes that Slack’s AI-driven search and summarization features address common challenges in information retrieval, saving time and boosting productivity across teams.
AI Innovations
1. Smell
Osmo has developed an AI-powered “scent teleportation” system that captures and recreates scents remotely without human involvement. Using AI and a Principal Odor Map (POM), the technology analyzes and transmits scent data for a molecular printer to synthesize, opening possibilities in virtual reality, therapeutic applications, and even remote scent sharing (Tech Radar).
2. Microsoft
Magentic-One, a newly introduced multi-agent AI system from Microsoft AutoGen, enables flexible, high-performance handling of open-ended web and file-based tasks, allowing an Orchestrator agent to coordinate specialized agents for activities like browsing, coding, and file navigation, with an emphasis on modularity, real-time updates, and safety protocols (Microsoft).
3. OpenAI
OpenAI’s new “Predicted Outputs” feature for GPT-4o and GPT-4o-mini significantly reduces latency by up to five times through speculative decoding, which skips over predictable content based on a reference string. This enhancement is especially useful for tasks like document editing, code refactoring, and other iterative text updates where response time is critical (Mark Tech Post).
4. NVIDIA
NVIDIA’s new AI Blueprint enables developers across industries to create visual AI agents that can analyze and summarize video and image data (NVIDIA).
NVIDIA has launched advanced AI and simulation tools, including Isaac Lab and Project GR00T workflows, to accelerate robot learning and humanoid development, enabling faster, high-quality robot control, dexterity, and environment interaction (NVIDIA).
5. Alphabet
Google AI Studio and Gemini API now offer “Grounding with Google Search,” enhancing Gemini models with real-time, accurate responses and in-line supporting links using Google Search (Google).
Google’s “Big Sleep” AI project has successfully identified a previously unknown, exploitable bug in SQLite, showing that large language models can uncover complex software vulnerabilities and assist in root-cause analysis, potentially strengthening defenses against cyber threats (PC Magazine).
Google’s new “Learn About” tool is an AI-powered, conversational learning aid that combines search and chat functionalities to offer personalized assistance for exploring complex topics. It allows users to input text, images, or files on any subject, then provides tailored, in-depth responses through interactive guides, images, and articles (ZDNet).
6. Anthropic
Claude AI’s new “Visual PDFs” feature allows users to analyze text, images, charts, and tables within PDFs, available through a paid subscription or API access (ZDNet).
Anthropic’s new Haiku 3.5 model offers high-speed processing, advanced tool-handling capabilities, and enhanced safety measures, making it popular for tasks needing efficiency and low oversight (Forbes).
7. Tencent
Tencent has launched Hunyuan-large, an open-source MoE language model. Hunyuan-large excels in multilingual NLP, code processing, and mathematical tasks, outperforming other open-source models like Llama3.1 (Pandaily).
8. AMD
AMD has released its first open-source 1 billion parameter language models, OLMo, optimized for general reasoning, chat capabilities, and responsible AI benchmarks, enabling efficient local deployment on AMD Ryzen AI PCs (AMD).
9. Kling AI
KLING now features “Custom Models,” allowing users to train unique video characters with consistent appearances across scenes by uploading 10–30 video clips (The Decoder).
10. Hume AI
Hume AI has launched a web-based app offering various AI-driven voice bots with distinct personalities for different types of interactions, such as quick answers, storytelling, and philosophical discussions (Tom’s Guide).
11. ByteDance
ByteDance’s new AI tool, X-Portrait 2, creates hyper-realistic videos from still images by capturing full facial movements rather than just tracking points (VentureBeat).
12. XPeng
At XPENG’s AI Day, the company unveiled the Kunpeng Super Electric System for rapid charging, the Turing AI Intelligent Driving System with L4 autonomous capabilities, the AIOS in-car interaction system, and the debut of the modular flying car and Iron humanoid robot (XPeng).
13. Nous
Nous has launched Nous Chat, a user-facing chatbot powered by its Hermes 3-70B LLM, a fine-tuned version of Meta’s Llama 3.1, offering a ChatGPT-like interface with prompt suggestions for knowledge, writing, and analysis (VentureBeat).
14. Video game model
Oasis is an AI-generated, open-world video game model that creates real-time gameplay using a transformer architecture, optimized for fast inference with Decart’s technology and designed to scale on Etched’s specialized hardware (GitHub).
15. 3D
Wonder Dynamics, now part of Autodesk, has launched Wonder Animation, a new AI tool that converts video sequences into 3D-animated scenes, allowing artists to film with various camera angles and cuts, then edit in a 3D space with full creative control (Autodesk).
Other Innovations
1. Drone delivery
Amazon has launched drone deliveries in parts of Phoenix for orders under five pounds, promising delivery within an hour during daylight and favorable weather conditions, following FAA approval for its MK30 drone to operate beyond visual line of sight (Engadget).
2. Cope with climate change
Jennifer Doudna, co-inventor of CRISPR, envisions a “revolution” in climate-adapted agriculture as CRISPR enables precise genetic edits that could make crops and animals more resilient to extreme weather. Her Innovation Genomics Institute (IGI) is working on drought-tolerant rice and low-methane cattle, as CRISPR’s fine-tuning offers an alternative to traditional genetic modification with fewer unintended changes (MIT Technology Review).
