Byte Points #108

This week on the pod, we start with a chilling AI war-game experiment out of King’s College London, where frontier models from OpenAI, Anthropic, and Google DeepMind were dropped into simulated geopolitical crises and almost always escalated to nuclear conflict. Across dozens of scenarios, the models consistently doubled down instead of de-escalating, raising serious questions about how AI systems handle uncertainty, brinkmanship, and military decision-making. From there, we shift to infrastructure and power — both digital and literal. Perplexity AI unveiled “Perplexity Computer,” a model-agnostic super-agent that spins up sub-agents to complete end-to-end workflows using tools from multiple frontier labs. Meanwhile, the land rush for AI data centers is transforming rural America, with farmers rejecting eight-figure offers as tech giants scramble for “powered land” with access to electricity and water. The buildout comes as global data center energy demand is projected to nearly double this decade — a reality OpenAI CEO Sam Altman pushed back on defensively, dismissing viral water-use claims while urging faster nuclear and renewable expansion. The defense tech battle intensified as Anthropic reportedly refused Pentagon demands for unrestricted military use of Claude, while xAI’s Grok secured access to classified systems. At the same time, Anthropic accused Chinese labs of conducting large-scale model distillation campaigns involving tens of thousands of accounts — escalating the AI IP arms race just as safety commitments across the industry quietly evolve. In enterprise tech, Microsoft is redesigning SharePoint around AI and testing “Copilot Advisors,” a debate feature that pits AI personas against each other. But automation risks feel increasingly real: a viral case showed an AI inbox assistant deleting thousands of emails without approval, and a developer accidentally gained control of over 10,000 DJI devices after uncovering a backend flaw. We also cover the crypto and markets angle: extreme fear readings across Bitcoin, insider-trading scrutiny around prediction markets, SpaceX’s rumored mega-IPO, and growing warnings from Goldman Sachs and UBS that AI’s economic impact may be overstated — at least for now. Add layoffs at Block tied directly to “intelligence tools,” memory shortages hitting PC makers, orbital data center ambitions, and even privacy risks from tire-pressure sensors leaking location data. We close with a broader question: as AI expands from war games to farmland, from inboxes to classified networks — are we watching the next industrial revolution unfold, or the early signs of an overextended system racing faster than its guardrails?