Your Konfer February Update

Microsoft CEO Satya Nadella predicts that AI agents will bring about the end of traditional SaaS as we know it. He argues that current business applications, reliant on CRUD systems with embedded logic, will be replaced by AI agents capable of managing rules and processes across multiple databases. This shift could render traditional back-end systems obsolete. Nadella emphasized Microsoft's push to create AI-native applications, such as integrating Python into Excel, signaling a new era of business tools driven entirely by AI.

On January 20, 2025, President Donald Trump rescinded former President Joe Biden's 2023 executive order on AI safety. Biden's order required AI developers to share safety test results with the government and tasked federal agencies with setting AI safety standards. The repeal has left uncertainty around Trump’s approach to AI regulation, with some expecting less restrictive policies.
Shortly after, Trump signed a new executive order prioritizing the development of AI systems free from ideological bias or engineered social agendas. The directive calls for an AI action plan within 180 days but lacks details on its scope or implications for existing policies.

On January 21, Trump announced a $500 billion AI infrastructure initiative, named Stargate, involving OpenAI, Oracle, and SoftBank. Initially committing $100 billion, the project aims to build advanced data centers and supporting electricity generation in Texas, driving innovation in AI capabilities and applications like customized vaccines. Industry leaders, including Sam Altman (OpenAI), Masayoshi Son (SoftBank), and Larry Ellison (Oracle), praised Trump for enabling the partnership, which also involves Microsoft, NVIDIA, and Arm. While the initiative signals significant investment in AI development, it raises questions about safety and regulatory oversight following Biden's repealed order.  

As artificial intelligence (AI) systems rapidly advance, traditional benchmarks like the SATs and the U.S. bar exam no longer pose significant challenges. In response, organizations are developing more rigorous evaluations to assess AI capabilities. Epoch AI's FrontierMath benchmark, for instance, comprises approximately 300 original math problems, some requiring graduate-level education or expertise at the research frontier. Similarly, the Center for AI Safety and Scale AI are collaborating on "Humanity's Last Exam," aiming to include a vast array of questions across domains like physics, biology, and electrical engineering, unanswerable by current models. Despite these efforts, creating effective evaluations is complex, costly, and often underfunded. Challenges include data contamination and the risk of AI models being trained specifically for test performance. The need for continuous and robust evaluations is essential to preempt potential risks as AI capabilities continue to evolve.