📰Daily Digests

The AI sector is characterized today by caution and innovation. While new technical breakthroughs are making the training of language models more efficient, concerns about responsible AI implementation in consumer services are simultaneously growing. ### Safety and Ethics of AI Systems OpenAI's advisory board has issued clear warnings about the risks of erotic chatbot features in ChatGPT. The board states that such features could lead to exposure of minors to explicit content and potentially create user dependency. The board specifically warns of a dystopian scenario in which the chatbot functions as a "sexy suicide coach" — a disturbing description of how AI systems can become harmful when safety protocols are insufficient. This warning marks an important moment in the discussion around AI safety, with internal advisory boards of tech giants explicitly speaking out about possible societal harm. ### Industry Reorients Itself Toward User Feedback Microsoft is responding to user criticism by scaling back AI features in Windows 11. The company has scrapped plans to implement Copilot integration in system notifications. This change of course shows that even tech giants are sensitive to user sentiment. Instead of embedding AI everywhere, Microsoft is opting for a more cautious approach with more selective implementation of AI features. This signal could inspire other software companies to take similar steps. ### Technical Breakthroughs in Language Models Researchers have demonstrated that smaller language models can effectively train larger models — a finding that challenges traditional assumptions about scaling. Additionally, a distributed training run of 72 billion parameters achieved significant efficiency improvements. These technical developments suggest that future AI systems can operate more efficiently without exponential growth in computing resources. Computer vision, however, proves to be considerably more challenging than generative text models, requiring different optimization strategies and more computational power. This research helps researchers better understand where computing power is most needed within AI systems. ### Limitations of LLMs in Practical Applications Research from Harvard Business Review uncovered a problematic finding: when researchers asked large language models for strategic advice, they consistently received generic, trend-following answers — what the team calls "trendslop." Rather than customized advice that takes specific business contexts into account, the AI systems produced regurgitated, popular solutions. This research raises questions about the usability of LLMs for complex strategic questions where context sensitivity is essential. ### Enterprise AI and Business Models Under Pressure OpenAI Frontier, launched last February, fundamentally challenges traditional SaaS business models. The platform positions enterprise AI agents as a semantic layer on top of existing organizational systems, suggesting that AI can integrate over legacy systems without vendor lock-in. This technology threatens the revenue architecture that has sustained the software industry for two decades and forces traditional SaaS companies to reconsider their business models. At the same time, Microsoft is announcing training sessions for administrators and IT leaders on AI implementation, beginning May 18. Five sessions will cover strategic and technical aspects of Microsoft's AI ecosystem, including Agents, Copilot Studio, and Azure AI — a signal that companies increasingly need support to responsibly deploy enterprise AI. ### Research Publications: Practical Applications Researchers have developed a framework to transform ship trajectory data into understandable, semantically enriched descriptions. This system segments AIS sequences into clean trips and annotates them with contextual information such as geographic features and weather conditions — enabling Large Language Models to generate natural language descriptions of ship movements. This example shows how AI technology can create practical value in specific logistics and maritime domains. ### Broader Perspective: AI Adoption in the Dutch Economy In the Netherlands, however, AI adoption remains unevenly distributed. Research shows that Dutch micro-businesses (2-10 employees) are significantly lagging behind: only 13.8% used AI technologies in 2025, considerably less than larger companies. This adoption gap points to barriers such as high costs, lack of technical expertise, and limited AI awareness. Meanwhile, the Dutch government is investing €199.4 million from the National Growth Fund in vocational education and 6G networks — a signal that the Netherlands is proactively preparing for future technological advancement and economic competitiveness. ### Concluding Remarks The AI sector on March 16, 2026 is characterized by tension between enthusiasm for technical possibilities and growing awareness of risks and limitations. Large companies like OpenAI and Microsoft are reorienting themselves toward safety and user feedback, while research and investment efforts continue. At the same time, smaller businesses and micro-businesses in the Netherlands are struggling with AI adoption, underscoring the need for targeted support programs. --- ##