AI for Science
Highlights
1. MIT researchers’ graph-based AI model integrates generative knowledge extraction and category theory to enable novel material designs.
2. OpenAI plans to launch “Operator,” an autonomous AI agent capable of independently controlling computers.
3. Microsoft introduced “adapted AI models” tailored for specific industries, enabling organizations to leverage pre-trained, fine-tuned AI capabilities.
4. Google has launched a dedicated Gemini AI app for iPhone.
5. TikTok has launched Symphony Creative Studio, an AI-powered tool suite to help advertisers and content creators quickly generate, transform, and scale video content.
6. Baidu has introduced enhanced text-to-image technology addressing “hallucination” issues, a no-code app builder for software creation, and AI-integrated smart glasses.
7. Stripe has introduced an agent toolkit enabling the integration of Stripe’s financial services into workflows powered by LLMs and function calling.
Innovation Insights
1. Why every company needs a futurist-in-residence (IDEO)
The futurist-in-residence role is vital for organizations aiming to navigate and shape the future by integrating foresight into their strategies and culture. This involves identifying subtle signals of change, mapping larger trends, and imagining diverse future scenarios to prepare for uncertainty and disruption. Tools like storytelling, creating future-focused artifacts, and scenario planning make abstract possibilities tangible and actionable for teams. By embedding futures-thinking through training and collaborative efforts, organizations can build resilience and agility to capitalize on emerging opportunities.
2. How GenAI reimagines supply chain management (BCG)
Generative AI is transforming supply chain management by simplifying interfaces, automating operations, and enabling actionable insights. It democratizes advanced tools, enhances decision-making, and fosters cross-functional collaboration through deeper automation and streamlined workflows. Successful adoption requires aligning GenAI with business goals, targeting high-impact areas, and rethinking workflows for seamless integration, while addressing talent gaps and building a robust partner ecosystem.
3. To help improve the accuracy of generative AI, add speed bumps (Ideas Made to Matter)
To address the challenges of bias and inaccuracies in generative AI, researchers recommend introducing “targeted friction,” such as cognitive and procedural speed bumps, to improve accuracy and reduce uncritical adoption. A study by MIT and Accenture demonstrated that moderate levels of friction, like using color-coded highlights to prompt users to review AI-generated content critically, balanced accuracy and efficiency effectively. This approach shifts users from automatic, fast decision-making (System 1) to more deliberate, thoughtful processing (System 2), without significantly hindering workflow speed.
AI Innovations
1. OpenAI
OpenAI’s ChatGPT for macOS now integrates with developer-focused apps like VS Code and Xcode, enabling context-aware code suggestions (TechCrunch).
AI companies like OpenAI are shifting from scaling pre-training with massive data and computing power to techniques like “test-time compute,” which enhance models during inference (Reuters).
OpenAI plans to launch “Operator,” an autonomous AI agent capable of independently controlling computers, as a research preview and developer tool in January 2025 (The Verge).
2. Anthropic
Anthropic has introduced a Prompt Improver and enhanced example management tools in the Anthropic Console, allowing developers to refine prompts using advanced techniques like chain-of-thought reasoning, example enrichment, and structured formatting (Anthropic).
3. Microsoft
Microsoft has introduced “adapted AI models” tailored for specific industries, enabling organizations to leverage pre-trained, fine-tuned AI capabilities through the Azure AI model catalog and Copilot Studio, with applications spanning healthcare, manufacturing, agriculture, finance, and automotive sectors (Microsoft).
4. Apple
Apple is reported to be expanding its smart home lineup with a wall-mounted smart display and a security camera, designed to integrate deeply with its HomeKit ecosystem, offering advanced AI capabilities (Forbes).
5. Alphabet
DeepMind’s new tool, Gemma Scope, uses sparse autoencoders to help researchers understand AI decision-making by visualizing internal model layers, potentially improving AI safety, control, and interpretability (MIT Technology Review).
Google DeepMind has made AlphaFold3’s code available for non-commercial academic use (Nature).
Google has launched a dedicated Gemini AI app for iPhone, offering a simple chat interface with text, voice, or camera input, and introducing the interactive Gemini Live feature for iOS users (The Verge).
6. Baidu
Baidu has expanded its AI offerings with enhanced text-to-image technology addressing “hallucination” issues, a no-code app builder for software creation, and AI-integrated smart glasses (Reuters).
7. Alibaba
Alibaba Cloud’s Qwen2.5-Coder, an open-source AI coding assistant supporting 92 programming languages, outperforms competitors like GPT-4 on key benchmarks (VentureBeat).
Alibaba has launched Accio, an AI-powered search engine for small businesses in Europe and the Americas, leveraging its Tongyi Qianwen language model to help users source wholesale products with consumer popularity and profit analyses, showing a 40% increase in purchase intent compared to traditional search engines (CNBC).
8. ByteDance
TikTok has launched Symphony Creative Studio, an AI-powered tool suite to help advertisers and content creators quickly generate, transform, and scale video content, featuring capabilities like auto-generated TikTok-ready videos, custom avatars, multilingual translation and dubbing, and AI-enhanced editing, designed to boost creative efficiency and ad performance (TikTok).
9. Mistral
AI startup Mistral has launched a customizable content moderation API, powered by its Ministral 8B model (TechCrunch).
10. Nous
Nous Research has launched the Forge Reasoning API Beta and Nous Chat, introducing enhanced reasoning capabilities via innovations like Monte Carlo Tree Search, Chain of Code, and Mixture of Agents, while providing a free platform for interacting with their advanced open-source Hermes 3 language model (Nous).
11. Stripe
Stripe has introduced an agent toolkit enabling the integration of Stripe’s financial services, such as invoicing, virtual card creation, and metered billing, into workflows powered by large language models (LLMs) and function calling, streamlining automation for tasks like payments, purchases, and usage-based billing (Stripe).
12. InVideo
InVideo has launched InVideo v3.0, a generative AI-powered feature allowing users to create and edit videos using text prompts in various styles (TechCrunch).
13. Robotics
MIT researchers developed LucidSim, a generative AI-powered system that creates realistic virtual environments to train robots for real-world tasks, achieving notably higher success rates than traditional simulation training (MIT Technology Review).
Johns Hopkins University researchers used imitation learning to train a surgical robot by analyzing videos of expert surgeons, enabling it to perform tasks like suturing and tissue manipulation with human-level skill, marking a breakthrough toward autonomous and precise robotic surgery (JHU).
14. Food ingredient quality
Singapore-based startup ProfilePrint uses AI and patented digital fingerprint technology to analyze ingredient quality rapidly, streamlining agribusiness supply chains (CNBC).
15. AI for science
Professor Markus Buehler’s graph-based AI model integrates generative knowledge extraction and category theory to uncover hidden connections across disciplines, enabling novel material designs inspired by biological systems, art, and music, with potential applications in sustainable materials, biomedical devices, and interdisciplinary innovation (MIT).
Other Innovations
1. Stretchable display
LG has unveiled a stretchable micro-LED display prototype that expands from 12 to 18 inches (Mashable).
2. Reinvent the wheel
South Korean Researchers have developed morphing wheels inspired by water droplets’ surface tension, which adapt to obstacles like stairs, curbs, and uneven terrain, offering potential applications for wheelchairs, unmanned delivery vehicles, and industrial robots requiring stable movement (Reuters).
