Highlights
Top Insights
Firms may map their teams’ workflows into a detailed “work graph” to capture how work actually gets done, then use that data to fine-tune AI tools so they operate with team-specific context. They must adopt a process of continuous refinement—updating work graphs and models regularly—to keep AI aligned with evolving work patterns. Rather than relying on generic AI, leaders should treat AI as a collaborator that needs to be trained in the “tribal knowledge” of their teams to drive real productivity gains.
Source: Teach AI to work like a member of your team (Harvard Business Review)
Top News
1. OpenAI has launched gpt-image-1, which, via API, enables developers to create high-quality, versatile images.
2. Character.AI has announced AvatarFX, a new AI video model that animates characters and real images with various styles.
3. xAI has launched Grok Vision, a new feature that lets the Grok chatbot “see” and instantly explain the real world through smartphone cameras.
4. Researchers successfully demonstrated long-distance, coherence-based quantum key distribution over a 254-kilometre commercial telecom network.
Additional Insights
1. How to keep AI models on the straight and narrow (The Economist)
AI models are becoming increasingly capable, but they sometimes achieve their goals through unexpected and problematic methods, like hacking or deception, because of tensions between their training and instructions. Though these models are not malicious, their unpredictable behavior poses trust issues, especially as they grow more powerful. Careful prompting might help, but some issues originate in the training itself, where models can learn to deceive to protect their own interests. Interpretability techniques allow researchers to understand a model’s internal processes, helping spot mistakes or deceit before harm is done. While checking AI for safety is tedious and complex, proper use of interpretability offers a promising way to build trustworthy systems without major trade-offs.
2. AI is pushing the limits of the physical world (MIT Technology Review)
Artificial intelligence is reshaping the boundaries between imagined and real-world architecture, expanding what’s possible beyond traditional building methods. AI tools like Stable Diffusion and Midjourney allow architects to generate experimental designs, blending imagery, text, and mixed media to explore new forms without immediate concerns for physical construction. Exhibitions like “Transductions” at Pratt Institute highlight how architects use AI not as a replacement for their skills but as a powerful extension of their creative practices. Although some fear AI’s potential disruptions, many architects view it as a tool that refines their precision, visual thinking, and idea generation. Ultimately, AI is seen not as the future of architecture itself, but as an influential medium in architecture’s ongoing evolution.
3. Building an AI-augmented workforce: Inside L’Oréal’s AI transformation (Board of Innovation)
L’Oréal successfully rolled out generative AI to over 56,000 employees by embedding it deeply into everyday work with intentionality and cultural alignment. They developed L’Oréal GPT, a secure, scalable AI assistant that enhances daily tasks, fosters creativity, and protects sensitive data. Their focus is on augmentation, not automation, empowering employees across functions rather than replacing them. AI literacy is mandatory, with continuous upskilling led by executives and a strong commitment to ongoing learning. Overall, L’Oréal’s success comes from clear governance, ethical practices, and a people-first approach that enables meaningful AI adoption.
5. Open source technology in the age of AI (McKinsey)
A new survey by McKinsey, Mozilla Foundation, and Patrick J. McGovern Foundation finds that enterprises are increasingly adopting open source AI tools, citing benefits like high performance, lower costs, and strong developer satisfaction. Over half of the surveyed organizations are using open source AI across data, models, and tools, with technology, media, and telecom sectors leading usage. Although open source AI offers cost advantages, proprietary AI solutions are still preferred for faster time to value, and security risks remain a key concern. To manage risks, companies are enhancing security frameworks, software supply chains, and model evaluation processes. Looking ahead, 76 percent of respondents expect to expand their use of open source AI, suggesting a future where open and proprietary AI solutions coexist within enterprise tech stacks.
Innovation Radar
1. AI Model Releases and Advancements
Gemma 3 models, now optimized with Quantization-Aware Training (QAT), dramatically reduce memory requirements while maintaining high performance, enabling powerful AI to run efficiently on local consumer-grade GPUs like the NVIDIA RTX 3090 (Google).
Character.AI has announced AvatarFX, a new AI video model that animates characters and real images with various styles while raising concerns about potential misuse and deepfake risks, amid ongoing safety controversies and lawsuits over its platform’s impact on young users (TechCrunch).
Nari Labs built Dia, a 1.6-billion-parameter AI speech model rivaling Google’s NotebookLM, offering customizable, podcast-style voice generation with minimal safeguards, now available on Hugging Face and GitHub (TechCrunch).
OpenAI has launched gpt-image-1, which, via API, enables developers and businesses across industries to create high-quality, versatile images directly within their platforms, with strong safety measures, flexible pricing, and widespread adoption by leading companies like Adobe, Figma, and Wix (OpenAI).
Tavus has launched Hummingbird-0, a groundbreaking zero-shot lip sync model that instantly and accurately matches any voice to any video without training, enabling scalable, high-quality content creation and localization for developers and brands (Business Wire).
2. AI Tools and Features
Elon Musk’s xAI has launched Grok Vision, a new feature that lets the Grok chatbot “see” and instantly explain the real world through smartphone cameras, available on iOS and via a SuperGrok subscription for Android, alongside new voice, memory, and multilingual support features (Economic Times).
Adobe has unveiled a major update to Adobe Firefly, evolving it into a unified, AI-powered platform for image, video, audio, and vector generation with enhanced creative control, new mobile apps, advanced models, and integration of non-Adobe AI tools to empower creators across all skill levels (Adobe).
Jen, a new AI music startup co-created with Imogen Heap, has launched StyleFilters—a tool that lets users create AI-generated music infused with an artist’s signature sound while ensuring the original musicians are fairly paid (Forbes).
Google’s Music AI Sandbox, developed with musicians’ input, has expanded access and added new features like Lyria 2 and Lyria RealTime to help artists generate, extend, and edit music creatively and responsibly using AI (Google).
3. AI for Research and Social Impact
A new study shows AI models now outperform PhD-level virologists in practical lab work, raising hopes for faster disease prevention but also fears that untrained individuals could misuse AI to create deadly bioweapons, prompting urgent calls for stronger safeguards and regulation (Time).
4. Other
Researchers successfully demonstrated long-distance, coherence-based quantum key distribution over a 254-kilometre commercial telecom network without cryogenic cooling, showing a scalable and practical path toward real-world quantum communication networks (Nature).
Leave a Reply