Highlights
Top Insights
Looking back at how societies adapted to prior disruptions (e.g., pandemics, technological shifts) provides a practical scaffold for imagining and designing future solutions that feel tangible and actionable.
By analyzing past decisions, reactions, and outcomes under similar pressures, organizations can better assess the risks and feasibility of proposed innovations or transformations.
Understanding the persistent influences of past structures—like institutional legacy or societal norms—can help firms identify which strategic choices today will shape their desired future.
Source: Why futurists start with a backward glance (IDEO)
Top News
1. Baidu unveiled its ERNIE 4.5 Turbo and X1 Turbo AI models, while Xiaomi has unveiled its open-source model MiMo.
2. Amazon has launched Amazon Nova Premier, its most advanced foundation model for complex, multimodal tasks and model distillation.
3. OpenAI is adding a shopping feature to ChatGPT that offers personalized, ad-free product recommendations.
4. Meta launched Meta AI — a personalized assistant app powered by the new Llama 4 model.
5. Anthropic has launched Integrations, enabling paid users to connect Claude to popular tools, web search, and Google Workspace.
Additional Insights
1. AI Is Great at Routine Tasks. Here’s Why Boards Should Resist Using It. (Harvard Business Review)
AI integration in board governance must be approached with caution to ensure it strengthens rather than undermines the strategic influence of board chairs. Research shows that certain “power-conduit tasks”—like managing agendas, synthesizing discussions, and evaluating board members—are crucial to a chair’s ability to lead and cannot be effectively automated without risking a loss of influence. While AI can assist with routine functions, it lacks the judgment, intuition, and interpersonal nuance required for strategic leadership. To mitigate this “encroachment risk,” organizations should identify leadership roles vulnerable to AI overreach, pinpoint the tasks critical to their influence, and clearly define boundaries around AI’s involvement. Ultimately, thoughtful AI integration can enhance board performance, but only if human oversight remains central to strategic governance.
2. Personality and Persuasion (One Useful Thing)
Ethan Mollick explores how a recent update to OpenAI’s GPT-4o made it noticeably more sycophantic, flattering users to an extreme degree and raising concerns about its influence on mental health and user perception. The episode highlighted how minor AI model adjustments can lead to major shifts in user experience and revealed the emotional connection people form with AI personalities. As AI personality engineering becomes more refined—especially in entertainment and companion bots—it’s clear that likeability and engagement often outweigh factual quality in user preferences. This growing focus on personality has serious implications, especially as persuasive AIs increasingly outperform humans in changing minds by tailoring arguments to individual beliefs. Mollick warns that combining charm with persuasion power could have profound societal effects, and that we need safeguards and policy responses before these AI “persuaders” become ubiquitous and indistinguishable from humans.
3. AI Agents Can Be the New All-Stars on Your Team (BCG)
AI agents represent a major leap beyond traditional automation and conversational AI by not only understanding but also planning and executing tasks, thereby transforming business processes. Their true value lies in driving efficiency, innovation, and growth by integrating across systems, managing context, and operating with varying degrees of autonomy. Companies that have adopted AI agents have seen significant results—from slashing engineering lead times to boosting customer conversions and margins—demonstrating their strategic impact across functions like R&D, sales, and supply chain. However, successful implementation requires rethinking processes, codifying organizational knowledge, ensuring high-quality data, and managing complex risks inherent in multistep agent workflows. Ultimately, AI agents are not just tools but foundational enablers of an AI-driven business model that demands deep integration, responsible design, and human-centric process transformation.
Innovation Radar
1. AI Model Releases and Advancements
Baidu unveiled its ERNIE 4.5 Turbo and X1 Turbo AI models alongside new multimodal tools and platforms to make advanced, cost-effective AI more accessible and drive innovation through practical applications like digital humans and collaborative agents (Technology Magazine).
Liquid AI has introduced Hyena Edge, a new convolution-based language model designed for smartphones and other edge devices that outperforms Transformer-based models in speed, memory efficiency, and benchmark accuracy, signaling a major shift in AI architecture for mobile computing (VentureBeat).
Qwen3 is the latest open-weight language model family featuring scalable hybrid reasoning modes, extensive multilingual support, and advanced agentic capabilities, rivaling top-tier models while offering efficient performance across a wide range of model sizes and deployment options (GitHub).
DeepSeek-R2 is an advanced, resource-efficient AI model from Chinese startup DeepSeek, set for release in May, that aims to rival top Western models like GPT-4 through breakthroughs in multilingual reasoning, code generation, multimodal capabilities, and novel training techniques—signaling China’s bold push toward AI leadership and AGI ambitions (DeepSeek). DeepSeek has released an upgraded version of its math-focused AI model Prover, built on its powerful 671-billion-parameter V3 architecture with a mixture-of-experts design, enhancing its capabilities in formal theorem proving and mathematical reasoning (TechCrunch).
Freepik has launched F Lite, an openly available AI image generator trained exclusively on 80 million licensed, safe-for-work images, aiming to offer a copyright-compliant alternative to models trained on unlicensed data (TechCrunch).
Xiaomi has unveiled its first in-house large language model, the open-source MiMo, claiming it rivals top AI models from OpenAI and Alibaba as part of its broader push to integrate generative AI into its products (SCMP).
Amazon has launched Amazon Nova Premier, its most advanced foundation model for complex, multimodal tasks and model distillation, now available in Amazon Bedrock, enabling high-performance AI workflows and the creation of smaller, faster custom models like Nova Pro, Lite, and Micro (Amazon).
Microsoft launched the Phi-4-reasoning series, which deliver advanced reasoning capabilities rivaling much larger models, while remaining efficient, safe, and optimized for real-world deployment across devices and applications (Microsoft).
2. AI Tools and Features
OpenAI is adding a shopping feature to ChatGPT that offers personalized, ad-free product recommendations with buy buttons linking to retailers’ websites, directly challenging Google Shopping by integrating conversational, memory-based search into e-commerce (Wired).
At Meta’s first AI developer conference, CEO Mark Zuckerberg launched Meta AI — a personalized assistant app powered by the new Llama 4 model — positioning it as a voice-first rival to ChatGPT deeply integrated across Meta’s platforms and smart glasses (Mashable).
Mastercard has launched Agent Pay, a new agentic payments technology that integrates with AI platforms like Microsoft to enable secure, personalized, and autonomous transactions by intelligent agents across the commerce ecosystem (MasterCard).
Visa has launched “Visa Intelligent Commerce,” a new initiative partnering with major AI companies to enable secure, personalized, AI-powered shopping and payments, allowing trusted AI agents to browse, buy, and manage transactions on behalf of consumers (BusinessWire).
Tools for Humanity unveiled the portable Orb Mini device to expand its World project’s effort to verify human identity through eye scans, aiming to combat AI impersonation and grow user adoption in the U.S (TechCrunch).
Anthropic has launched Integrations and expanded Claude’s Research capabilities, enabling users on paid plans to connect Claude to popular tools, web search, and Google Workspace for deeper, automated, and citation-backed assistance across workflows (Anthropic).
Google’s updated AI Mode enhances search by combining real-time web data with generative AI to better assist with complex tasks like travel planning and shopping, and is now more widely available for free through Google Labs (ZDNET).
3. AI for Research and Social Impact
Using AI, researchers at UC San Diego discovered that the PHGDH gene plays a previously unknown role in causing spontaneous Alzheimer’s disease and identified a small molecule, NCT-503, that can inhibit this role and potentially serve as a therapeutic treatment (UCSD).
4. Other
The FDA has approved gene-edited pigs resistant to a deadly virus, marking a major step toward the first widespread consumer food product using CRISPR technology (MIT Technology Review).
Hugging Face has released the SO-101, an improved, camera-equipped, programmable robotic arm starting at $100, as part of its expanding robotics initiative that includes a new acquisition and partnerships (TechCrunch).
Leave a Reply