AI shopping assistant
Highlights
Insights
1. GenAI shows promise in helping with strategic management tasks, but falls short in human-like intuition and sophisticated analysis.
2. Scientific AI, a specialized approach leveraging AI for hypothesis generation and testing, has the potential to revolutionize R&D.
3. AI mistakes are different than human mistakes, making them less predictable and harder to trust.
News
1. NVIDIA has unveiled a new AI Blueprint for retail shopping assistants, leveraging generative AI and 3D visualization technologies to enhance customer experiences.
2. Google Automotive AI Agent enables automakers to build in-car assistants.
3. OpenAI has developed a specialized AI model, GPT-4b micro, designed to engineer proteins for converting regular cells into stem cells.
4. Sky-T1-32B-Preview is an open-source reasoning model performing competitively with o1-preview in math and coding benchmarks.
Innovation Insights
1. CES 2025: What We Saw—and Wish We Had Seen (IDEO)
The 2025 CES showcased a whirlwind of AI-driven innovations, including smart glasses, creator tools, and immersive media experiences, but highlighted a gap between tech designed to enhance lives and those pushing users to reorganize around products. Panel discussions emphasized the need for meaningful, intuitive technology, particularly for children, who value human connection and usability over novelty, a principle that extends across demographics. Smart glasses stood out for their growing versatility, though their audio integration remains underdeveloped, while creator tools made professional-grade capabilities accessible to non-experts, underscoring the importance of storytelling with heart. The event also reflected a rise in location-based entertainment, like immersive fandom experiences, and a nostalgic resurgence of simpler storytelling formats like webtoons and anime. However, there was a notable absence of innovative solutions for pressing human challenges, such as women’s health and aging care, signaling opportunities for future breakthroughs that prioritize dignity, connection, and joy.
2. Can GenAI do your next strategy task? Not yet. (California Management Review)
GenAI shows promise in assisting with strategic management tasks, particularly in areas requiring data synthesis and limited reasoning, such as market research. However, its ability to independently handle complex, multi-step tasks like strategic scenario planning or M&A due diligence remains limited, necessitating a human-in-the-loop to ensure quality and accuracy. Recent advances, like OpenAI’s o1-preview model, have improved performance significantly in tasks involving multistep reasoning and context-dependence but still fall short in mimicking human behavioral intuition and fully automating sophisticated analyses. Despite these limitations, businesses should consider leveraging these tools now, as their capabilities already provide substantial productivity gains, while remaining aware of the evolving landscape of AI technology.
3. Scientific AI: Unlocking the next frontier of R&D productivity (McKinsey)
Scientific AI, a specialized approach leveraging AI for hypothesis generation and testing, has the potential to revolutionize R&D by accelerating discovery and creating new solutions across various scientific disciplines. Unlike general AI tools, Scientific AI integrates proprietary data and cross-disciplinary insights, addressing long-standing silos in research and enabling advancements in fields like drug discovery, materials science, and protein engineering. This iterative process of AI-driven hypothesis testing and refinement is bolstered by foundation models and active-learning loops, unlocking both productivity gains and transformative innovation. However, successful deployment requires a comprehensive strategy encompassing infrastructure, talent, and scaling plans, emphasizing gradual adoption aligned with strategic priorities to realize its full potential.
4. GenAI Isn’t Just a Headache for Risk and Compliance. It’s an Enabler (BCG)
GenAI is creating challenges for risk and compliance (R&C) teams due to privacy, regulatory, and ethical concerns. Instead of resisting the technology, R&C teams should proactively adopt GenAI to enhance their efficiency, improve risk management, and align compliance efforts with organizational growth goals. GenAI enables automation of repetitive tasks, enhances real-time risk mitigation, and supports strategic objectives by improving operational efficiency, fraud detection, healthcare R&D, personalized marketing, and energy project compliance. Responsible AI (RAI) practices are essential to prevent ethical, legal, and privacy violations, ensuring the ethical and sustainable adoption of GenAI. By embracing GenAI and RAI, R&C teams can transform into strategic partners, fostering innovation while ensuring ethical, legal, and regulatory compliance.
5. AI Mistakes Are Very Different Than Human Mistakes (IEEE Spectrum)
Human mistakes are predictable, often tied to fatigue or gaps in knowledge, and mitigated by established systems like double-checking and accountability practices. In contrast, AI systems like large language models (LLMs) make mistakes that are random, bizarre, and unaccompanied by signs of ignorance, making them less predictable and harder to trust in complex tasks. To address these differences, researchers propose two approaches: making AI errors more human-like and developing specialized systems to catch and correct AI-specific mistakes. Understanding the peculiarities of AI errors, some of which mimic human biases while others are entirely unique, is essential for safely integrating AI into decision-making processes.
AI Innovations
1. Alphabet
Google Cloud has unveiled its Automotive AI Agent, powered by Gemini with Vertex AI, enabling automakers like Mercedes-Benz to create advanced, multilingual, and conversational in-car assistants, debuting in the new Mercedes-Benz CLA’s MBUX Virtual Assistant later this year (Google).
2. Microsoft
MatterGen, a generative AI tool introduced by Microsoft researchers, revolutionizes materials design by directly generating novel, stable materials tailored to specific applications, significantly outperforming traditional screening methods in efficiency and scope (Microsoft).
3. OpenAI
OpenAI’s Economic Blueprint outlines strategies to ensure the U.S. leads in AI innovation, fosters equitable access to its benefits, and drives national economic growth. It emphasizes collaboration with policymakers to create common-sense regulations that safeguard public safety while encouraging innovation, reindustrialization, and global competitiveness (OpenAI).
OpenAI has developed a specialized AI model, GPT-4b micro, designed to engineer proteins for converting regular cells into stem cells, achieving more than a 50-fold efficiency improvement in modifying Yamanaka factors, which rejuvenate cells (MIT Technology Review).
OpenAI’s new “Tasks” feature for ChatGPT, available to paying users, introduces scheduled and recurring task automation, signaling a step toward integrating traditional AI assistant functionalities with advanced agentic AI capabilities (Ars Technica).
4. NVIDIA
NVIDIA has unveiled a new AI Blueprint for retail shopping assistants, leveraging generative AI and 3D visualization technologies to enhance customer experiences by enabling personalized, intelligent, and visually accurate interactions both online and in-store (NVIDIA).
NVIDIA has partnered with leading organizations like IQVIA, Illumina, Mayo Clinic, and Arc Institute to revolutionize healthcare and life sciences by leveraging AI and accelerated computing. These collaborations aim to enhance drug discovery, genomic research, clinical trials, and personalized medicine through AI-powered tools, robotics, and advanced computational platforms (NVIDIA).
5. Mistral
Codestral 25.01 is an advanced coding model optimized for fast, precise coding tasks, outperforming competitors in Python coding benchmarks with an 86.6% HumanEval score (VentureBeat).
6. DeepSeek
DeepSeek introduced a free iOS app, based on its open-source DeepSeek-V3 model (Tech in Asia).
7. MiniMax
MiniMax has unveiled three advanced models—MiniMax-Text-01, MiniMax-VL-01, and T2A-01-HD—claimed to rival top global AI systems in text, multimodal, and audio generation (TechCrunch).
8. Open-source
Sky-T1-32B-Preview, an open-source reasoning model performing competitively with o1-preview in math and coding benchmarks, was trained for under $450, demonstrating affordable, efficient high-level reasoning capabilities while providing open-access data, code, and model weights for community collaboration (NovaSky).
9. Video
Luma Labs’ new Ray2 video model, integrated with the Dream Machine platform, delivers highly realistic, physics-aware video generation, surpassing competitors like Runway Gen-3 and OpenAI Sora, and can produce 10-second high-resolution clips from text or image prompts (Tom’s Guide).
10. Translation
The SEAMLESSM4T system, a groundbreaking multilingual and multimodal translation model, unifies speech and text translation across numerous languages, outperforming state-of-the-art cascaded systems in quality, robustness, and inclusivity, while being publicly available for non-commercial research (Nature).
Other Innovations
1. Neuralink
Neuralink plans to expand its human trials in 2025, refining its brain implant technology for better control, faster calibration, and applications like robotic arm usage, though commercialization remains years away (MIT Technology Review).
2. Cancer treatment
Scientists have developed a novel cancer treatment using engineered viruses to disguise tumors as pig tissue, triggering an immune response akin to organ rejection, showing promise in early trials but requiring further testing to confirm efficacy and safety (Nature).
